Stripe Agentic Commerce vs PayPal Agentic Commerce

Cart

1. ai-native userMy agent can create and update a cart or order draft through a documented protocol call

   weight 3 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce~partialclaimed7/10

   Stripe's ACP-style custom checkout documents an explicit reverse-API contract where sellers implement create/update/confirm endpoints and sessions move through documented statuses (incomplete | ready_for_payment | requires_escalation | processing | completed | canceled), directly matching cart/order-draft creation and update via a documented protocol call; the embedded OCA flow also covers 'cart management' end-to-end. However this is gated behind private preview, US-only availability, and a waitlist, with no independent/hands-on evidence of a successful cart-update call in the pack. Missing for 10: independent/hands-on verification of a live create/update call succeeding, and confirmation the feature is generally available rather than waitlisted preview.

   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullclaimed8/10

   PayPal documents a concrete, callable cart protocol: POST /merchant-cart createcart to create a cart from a PayPalCart object and getcart to retrieve/update cart status, validation issues, and totals, explicitly invoked by AI agents/PayPal Shopping Cart during checkout; the agent toolkit also exposes create_order and related order-management tools callable by AI agents via MCP/LangChain/etc. Missing for 10: no independent hands-on developer report confirming a live end-to-end agent-driven cart creation/update flow beyond PayPal's own docs, and cart update (vs. only create/get) isn't explicitly documented as a separate call.

   • [claimed-docs] “Creates a new cart with the provided PayPalCart object. This is called by PayPal Shopping Cart when an AI agent or customer initiates a chec…”
   • [claimed-docs] “Retrieves the current state of a cart. Used by PayPal Shopping Cart to check cart status, validation issues, and current totals.”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”

   proof ↗⚑ flag

Checkout

1. ai-native userMy agent can complete a purchase end-to-end — checkout session, delegated payment credential, confirmed order

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Stripe documents the full end-to-end agent checkout: sellers publish catalogs via ACS, agents complete checkout via ACP/UCP or embedded flows, shared payment tokens serve as delegated payment credentials with scoped usage limits, and completion is confirmed via checkout.session.completed events with agent attribution; runtime probes confirm the MCP server, Link CLI, and MPP infra are live. missing for 10: independent hands-on end-to-end purchase report (community evidence is speculative/skeptical, not a confirmed successful transaction), and several flows (ACS custom checkout, OCA embedding) remain private-preview/waitlisted rather than generally available.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases.”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   Docs describe a real end-to-end path: agent-initiated checkout session (ACP requestCheckout / UCP checkout-session), a delegated one-time-use payment credential (Braintree nonce / Google Pay token), and order confirmation flowing into the merchant's existing order system, backed by a cart API and agent toolkit tools (create_order, pay_order). However this is gated behind Braintree accounts, PayPal's form-based approval, Google's separate merchant enablement, and Store Sync's narrow scope (US, physical goods only) — all from first-party docs with no independent or hands-on confirmation of a completed purchase. Missing for 10: independent/hands-on evidence of an actual completed agent purchase, broader (non-gated) merchant eligibility, and real-world confirmation beyond sandbox/spec descriptions.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”

   proof ↗⚑ flag

Embedded surfaces

1. merchantBuyers can complete purchases of my products inside chat and assistant surfaces (ChatGPT, Claude, Gemini, Copilot and peers)

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Stripe's ACP/ACS clearly enables in-chat browsing and checkout, with concrete evidence tied to ChatGPT (OpenAI co-developed ACP, 'first AI platform to implement ACP with ChatGPT', and merchant onboarding/CSV catalog/checkout-session flows), plus SPT tokenization and human-approval gates. However, evidence names only ChatGPT as a live surface — there is no documentation or proof of live integration with Claude, Gemini, or Copilot — and the embedded/agent-as-intermediary checkout mode is explicitly gated as 'private preview' and limited to 'US, Canada, and select European countries,' with ChatGPT participation requiring an application rather than being open by default. Community threads raise trust/incentive concerns but don't concretely contradict that checkout works for ChatGPT. missing for 10: evidence of live checkout support in Claude/Gemini/Copilot, and proof the private-preview gating has lifted to general availability.

   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”
   • [claimed-docs] “"ACP is open source and community-designed under the Apache 2.0 license. Businesses can implement the specification to transact with any AI …”
   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   PayPal offers real infrastructure for in-chat checkout: ACP for ChatGPT (via Braintree delegated tokens), UCP for Google AI Mode/Gemini (via Google Pay tokens), and Store Sync for broader catalog/agent discovery, plus an MCP server and agent toolkit. However, coverage is fragmented and gated — Store Sync is US-only, physical-goods-only, and access is form-gated with manual onboarding; ACP/UCP require an existing Braintree relationship or separate Google merchant registration; no evidence of Claude or Copilot checkout support is present. Missing for 10: broader platform coverage beyond ChatGPT/Google (Claude, Copilot), self-serve enablement without form gating, and independent/hands-on confirmation of live buyer transactions completing in-chat.

   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”

   proof ↗⚑ flag

Protocol conformance

1. developerBuild against a versioned commerce protocol with a published schema and conformance or validation tooling

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialprobed6/10

   Stripe documents an open, community-governed protocol (ACP, Apache 2.0, published spec at agenticcommerce.dev) and, for the sibling Machine Payments Protocol, an explicit conformance CLI ('mppx validate' tests discovery, challenge formats, error handling, and the full payment flow) plus a public spec site (mpp.dev) confirmed live by probe. However, there's no evidence of a formally versioned schema (e.g., semver, changelog) for ACP itself, no OpenAPI/JSON schema found (probe explicitly 404s for openapi.json/swagger.json), and no ACP-specific validator analogous to mppx validate — conformance tooling is documented only for MPP, not the core commerce checkout protocol. missing for 10: versioned ACP schema/spec artifact, ACP-specific conformance/validation tool, published machine-readable OpenAPI schema.

   • [claimed-docs] “"ACP is open source and community-designed under the Apache 2.0 license. Businesses can implement the specification to transact with any AI …”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed4/10

   PayPal does version its agentic-commerce cart API (v1 createcart/getcart) and provides catalog-feed validation tooling for Store Sync ('a validated product catalog file' against Google/OpenAI ACP feed specs), and it interoperates with external versioned protocols (ACP, UCP). But there's no evidence of a published, standalone schema (e.g., OpenAPI/JSON Schema) for PayPal's own agentic-commerce protocol nor a formal conformance/test suite for developers to validate protocol implementations beyond feed-file field checks. Missing for 10: published protocol schema/spec document, dedicated conformance or validation test suite for the cart/checkout API (not just catalog feed format), and independent evidence of developers using such tooling.

   • [claimed-docs] “Creates a new cart with the provided PayPalCart object. This is called by PayPal Shopping Cart when an AI agent or customer initiates a chec…”
   • [claimed-docs] “Retrieves the current state of a cart. Used by PayPal Shopping Cart to check cart status, validation issues, and current totals.”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”

   proof ↗⚑ flag

Agent search

1. ai-native userMy agent can programmatically search products or services across merchants — keylessly or with self-service credentials, no private partnership required

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed4/10

   Stripe does expose some keyless/self-service discovery paths — the Link CLI lets an agent 'Search products and complete payments with UCP' using a customer-controlled wallet with self-service credentials (no merchant partnership needed on the buyer side), and Stripe Directory helps agents find external providers. But the flagship agentic-commerce discovery/checkout path (ACS/OCA) explicitly requires a private, seller-initiated relationship — 'Only a seller can initiate an OCA request' and 'An OCA is a required connection between your agent and a seller that enables agentic commerce flows' — and is gated as 'private preview' in limited geographies, meaning broad cross-merchant search without any partnership is not generally available. missing for 10: universal, partnership-free product search across arbitrary merchants (most flows require seller-approved OCA or private-preview access), independent/hands-on confirmation that Link CLI's UCP search actually returns cross-merchant results without any seller-side integration.

   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [claimed-docs] “Stripe Directory helps developers and AI agents find the best external providers for a task and follow the best supported path to provision …”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”

   proof ↗⚑ flag

   PayPal Agentic Commerce!disputedcontradicted3/10

   PayPal frames agentic commerce as 'connect once to reach many platforms', but access to Store Sync/agentic commerce services is explicitly form-gated ('merchants must complete a form... to request access... PayPal AI team will follow up to guide onboarding'), Agent Ready requires an existing Braintree merchant relationship, and UCP requires separate Google merchant registration — none of this is keyless or self-service, and runtime probing confirms the live MCP server itself demands OAuth Bearer authentication (401 + www-authenticate) rather than allowing keyless discovery calls. This contradicts the story's requirement of keyless or self-service, no-partnership product search across merchants. Missing for 10: keyless/anonymous cross-merchant product search, a genuinely self-serve onboarding path without a sales/approval form, and evidence of catalog search working without a pre-existing Braintree/PayPal merchant relationship.

   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”

   proof ↗⚑ flag

Machine readable offers

1. developerRead machine-readable offer data — price, availability, variants, checkout links — for any listed product or service

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed6/10

   Stripe documents a seller-side catalog feed (CSV) containing product/variant rows with price and availability fields, plus checkout completion via shared payment tokens and delegated checkout webhooks, and agent-builder docs describe 'product feed ingestion' so agents can present offers and checkout. However, the mechanism is a seller-pushed feed into Stripe's index rather than a documented universal read/query API that any developer/agent can call to fetch offer data for an arbitrary listed product, and it's gated behind private preview, onboarding agreements (OCA), and US/Canada/EU-only availability. Missing for 10: a documented public read/query endpoint or protocol response schema for agents to pull offer data on demand, and independent/hands-on confirmation of catalog read access working end-to-end.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”
   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed5/10

   PayPal's Store Sync catalog feed spec and agent-toolkit's list_product/show_product_details tools do define machine-readable offer fields (id, title, price, availability, description, image_link, link — with each row representing a product variant) and cart-read endpoints expose price/checkout totals, so the core data model the story asks for exists. But this is a gated, merchant-onboarding pathway (form approval required, US-only physical goods, Braintree/PayPal-account prerequisite) rather than an open read surface for 'any listed product,' and there's no evidence of a simple public GET-offer endpoint or checkout-link retrieval independent of the full agentic-commerce integration. Missing for 10: an open/self-serve read API (vs form-gated onboarding), broader product/geography scope, explicit example of returning a checkout link, independent/hands-on corroboration of reading catalog data.

   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”

   proof ↗⚑ flag

Merchant feeds

1. merchantPublish my product or service catalog into agent surfaces through a documented feed or import API so AI agents can discover what I sell

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed9/10

   Stripe documents a concrete, detailed catalog-publishing mechanism for merchants: a CSV feed format uploaded via the v2 Product Catalog Import API to a presigned URL, with defined cadence (daily product data, 15-min inventory/pricing), explicit replace-mode deletion semantics, import-result webhooks with per-row error reporting, and required restricted-key permissions — all under the self-serve ACS onboarding flow described in the for-sellers docs. This directly matches the story of publishing a catalog into agent surfaces via a documented feed/import API, going well beyond a vague claim into an operational spec merchants can implement against. missing for 10: independent/hands-on merchant confirmation of the import pipeline working end-to-end, and the feature is still gated to private preview / limited geographies (US, Canada, select European countries) rather than universally GA.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration.”
   • [claimed-docs] “It lets you share product, price, and availability information with agents while minimizing changes to your existing commerce systems.”
   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed6/10

   PayPal documents a real catalog-publishing path (Store Sync with feed specs like Google Product Feed/OpenAI ACP Product Feed, CSV/TSV/PSV, field requirements) that explicitly targets AI-agent discovery, plus partner-based catalog connections (Wix, Cymbio, BigCommerce/Feedonomics, Shopware). However, access is form-gated (not self-serve), scope is narrow (US-only, physical goods, USD only), and eligibility requires an existing PayPal/Braintree merchant account with compatible setup. missing for 10: self-serve onboarding without a sales-gated form, broader product-type/geography coverage, independent merchant confirmation the feed import actually surfaces products in live agent surfaces (ChatGPT/Gemini) rather than only demo/spec documentation.

   • [claimed-docs] “Easy setup: Connect your product listings to PayPal's partners like Wix, Cymbio, Commerce (BigCommerce & Feedonomics), and Shopware to make …”
   • [claimed-docs] “It connects your product catalog and commerce API with PayPal's agentic commerce services, enabling AI agents to discover your products, cre…”
   • [claimed-docs] “By the end of this page, you will have a validated product catalog file that is ready to connect to Store Sync.”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”

   proof ↗⚑ flag

Service discovery

1. ai-native userMy agent can discover payable services and APIs through a published directory or discovery index

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed7/10

   Stripe publishes a live, machine-readable discovery surface: the 'Stripe Directory' explicitly helps 'developers and AI agents find the best external providers for a task and follow the best supported path to provision or use them' (docs-13), backed by a probed live well-known skills/services index (stripe-ac-probe-rt-3) and the open mpp.dev spec index for machine-payable APIs (stripe-ac-probe-rt-2), plus per-endpoint discovery via HTTP 402 challenges (docs-9/18/19). Missing for 10: concrete evidence that the Directory itself indexes specifically 'payable' services/APIs (vs. general providers) and an end-to-end example of an agent using the directory to find and pay for a service.

   • [claimed-docs] “Stripe Directory helps developers and AI agents find the best external providers for a task and follow the best supported path to provision …”
   • [probe] “PROBE runtime (recorded 2026-09-14): the machine-readable skills catalog is live at the documented keyless well-known URL — https://docs.str…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [claimed-docs] “x402 is a protocol for internet payments... your server returns an HTTP 402 response with payment details, including a Stripe deposit addres…”
   • [claimed-docs] “Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid …”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed5/10

   PayPal provides discovery mechanisms — Store Sync product-catalog feeds (Google Product Feed / OpenAI ACP Product Feed) that let AI agents 'discover your products' and a live MCP server discoverable via RFC 9728 well-known metadata — but there is no published cross-merchant directory or marketplace index of payable services/APIs; discovery is scoped to a single merchant's catalog after form-gated onboarding, and Store Sync itself is limited to US physical goods only. missing for 10: a cross-merchant/public discovery index of payable services or APIs, evidence of self-serve (non form-gated) discoverability, and support beyond US physical goods.

   • [claimed-docs] “It connects your product catalog and commerce API with PayPal's agentic commerce services, enabling AI agents to discover your products, cre…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [probe] “official MCP server documented at https://developer.paypal.com/ai-tools/mcp-server”

   proof ↗⚑ flag

Agent access

1. ai-native userPoint an agent at llms.txt or agent-oriented docs

   weight 2 · round drawn
   Stripe Agentic Commerce✓fullprobed9/10

   Stripe confirms a live llms.txt (HTTP 200 probe) and extensive agent-oriented markdown docs (.md endpoints, skills.json well-known catalog, MCP server, link-cli --llms-full), all directly consumable by an agent pointed at these URLs. missing for 10: independent third-party confirmation that agents actually consume/parse these files effectively in the wild beyond Stripe's own probes.

   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.stripe.com/llms.txt # Stripe Documentation When installing Stripe packages, always check the npm r…”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/agentic-commerce.md # Agentic commerce Sell through agents, or embed commerce into your …”
   • [probe] “PROBE runtime (recorded 2026-09-14): the machine-readable skills catalog is live at the documented keyless well-known URL — https://docs.str…”
   • [claimed-docs] “Install instructions that help AI agents create more accurate Stripe integrations.”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullprobed9/10

   There's a confirmed live llms.txt at developer.paypal.com/llms.txt (HTTP 200) plus per-page .md agent-readable docs (e.g. agent-ready/overview.md), and the docs are explicitly agent-oriented (MCP server, agent toolkit, ACP/UCP guides) that an AI agent could consume directly. Missing for 10: no independent third-party report of an agent actually consuming llms.txt successfully in the wild, just first-party probes.

   • [probe] “PROBE llms.txt: HTTP 200 at https://developer.paypal.com/llms.txt # PayPal Developer Documentation > PayPal's developer documentation for i…”
   • [probe] “PROBE docs-md: HTTP 200 at https://developer.paypal.com/agent-ready/overview.md # Agent Ready overview (/agent-ready/overview) Agent Read…”
   • [probe] “official MCP server documented at https://developer.paypal.com/ai-tools/mcp-server”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”

   proof ↗⚑ flag

2. ai-native userRun the product headlessly / in CI for automation

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Stripe ships CLI/API primitives that could run headlessly — anonymous CI sandbox creation via `stripe sandbox create` (no account registration), an `mppx validate` conformance CLI for automated end-to-end testing of MPP integrations, and REST/MCP APIs that are inherently non-interactive. However, the core agentic-commerce and Link CLI wallet flows are explicitly gated by mandatory human approval ("Stripe requires human confirmation before it takes certain stripe_api_write actions"; "Customers approve each request on the Link website or in the mobile app"), which blocks fully unattended CI automation of actual payment execution. Missing for 10: explicit CI/pipeline documentation or examples, a documented way to bypass/automate the human-approval step for testing, and evidence of headless operation for the core checkout/SPT flows rather than just sandbox/dev tooling.

   • [claimed-docs] “Coding agents should install the Stripe CLI (npm i -g @stripe/cli) and run the command stripe sandbox create --help to provision an anonymou…”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   PayPal's agent toolkit (npm/PyPI packages) and REST APIs, plus the sandbox environment, can be scripted headlessly and invoked from CI pipelines via SDKs (LangChain, CrewAI, OpenAI Agents SDK, MCP) without a UI; webhooks and sandbox credentials support automated testing. However the remote MCP server requires interactive OAuth/PKCE authorization flow (dynamic client registration, bearer auth), which complicates pure headless/CI use, and there's no explicit CI/automation guide or example pipeline in the evidence. missing for 10: documented headless auth flow for CI (service-account/client-credentials for MCP), explicit CI/automation examples, and independent confirmation of unattended operation.

   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”

   proof ↗⚑ flag

3. ai-native userConnect an agent via an official MCP server

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed9/10

   Stripe documents and runtime-verifies a first-party hosted MCP server (mcp.stripe.com) with clear auth (OAuth 401 challenge observed live) and documented tools for AI agents to interact with the Stripe API, plus human-approval gating for sensitive actions. This is corroborated by both docs and a runtime probe confirming the endpoint is live and gated exactly as documented. Missing for 10: independent third-party (non-Stripe) hands-on report of connecting an agent to the MCP server in production use.

   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API”
   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API and search Stripe’s kno…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullprobed8/10

   PayPal documents and runtime-probes an official MCP server (developer.paypal.com/ai-tools/mcp-server) plus a published agent-toolkit npm/PyPI package explicitly listing MCP as a supported integration, and a live probe confirms the remote MCP endpoint (mcp.paypal.com/mcp) responds with proper OAuth-protected-resource metadata. missing for 10: independent/community hands-on confirmation of a successful end-to-end MCP client connection (beyond PayPal's own docs and the auth-handshake probe), and clearer documentation of the full tool list exposed over MCP.

   • [claimed-docs] “PayPal built an MCP server that lets merchants use natural language with their favorite MCP clients. This helps users complete business task…”
   • [probe] “official MCP server documented at https://developer.paypal.com/ai-tools/mcp-server”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”

   proof ↗⚑ flag

4. ai-native userUse an official CLI

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Stripe ships the official Link CLI (@stripe/link-cli, published on npm and confirmed live in a runtime probe, v0.19.2) for agent wallet access, plus the Stripe CLI for sandbox provisioning and mppx CLI for MPP conformance testing, all documented and agent-legible (--llms-full docs). missing for 10: no independent/community hands-on review of the CLI's usability, and multiple overlapping CLIs (stripe, link-cli, mppx) rather than one unified official CLI.

   • [claimed-docs] “Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to comple…”
   • [github] “The CLI can produce one of three credential types: A virtual card (PAN)... A Link Pay Token (LPT)... A Shared Payment Token (SPT)”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [probe] “official CLI documented at https://docs.stripe.com/agentic-commerce/link-cli”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”
   • [claimed-docs] “Coding agents should install the Stripe CLI (npm i -g @stripe/cli) and run the command stripe sandbox create --help to provision an anonymou…”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack documents PayPal's Agent Toolkit, MCP server, REST APIs, and SDKs, but nowhere mentions an official command-line interface (CLI) tool for developers or AI-native users to interact with PayPal's agentic commerce services. Since PayPal is a developer-facing platform with APIs and SDKs, a CLI would be a plausible product offering, but none is evidenced here.

   ⚑ flag

5. ai-native userDrive the product through a documented public API

   weight 3 · round drawn
   Stripe Agentic Commerce✓fullprobed9/10

   Stripe publishes extensive, concrete public API documentation for agentic commerce — ACS seller/agent endpoints, shared payment tokens, machine payments (MPP/x402), a hosted MCP server, and a CLI — all with runtime-verified live endpoints (401-gated MCP server, live mppx package, live skills.json, live link-cli npm package), giving AI-native users clear documented ways to drive the product programmatically. Missing for 10: a discoverable formal OpenAPI/swagger schema (probe-3 shows all candidate spec URLs 404) and independent third-party hands-on API integration reports beyond Stripe's own docs.

   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [probe] “PROBE runtime (recorded 2026-09-14): the machine-readable skills catalog is live at the documented keyless well-known URL — https://docs.str…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullprobed9/10

   PayPal publishes extensive, concrete developer documentation (REST API reference, Agent Toolkit, MCP server, agentic-commerce cart API, Store Sync, ACP/UCP guides) with runtime confirmation of a live MCP endpoint and published npm/PyPI SDK packages, giving AI-native developers a genuinely documented public API surface to build against. Missing for 10: some flows are form-gated/partner-gated (Braintree, Store Sync approval) rather than fully self-serve, slightly limiting pure API-driven access.

   • [claimed-docs] “PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using Amazon Bedrock, CrewAI, LangChain, Model Contex…”
   • [claimed-docs] “PayPal built an MCP server that lets merchants use natural language with their favorite MCP clients. This helps users complete business task…”
   • [claimed-docs] “Creates a new cart with the provided PayPalCart object. This is called by PayPal Shopping Cart when an AI agent or customer initiates a chec…”
   • [claimed-docs] “Retrieves the current state of a cart. Used by PayPal Shopping Cart to check cart status, validation issues, and current totals.”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”

   proof ↗⚑ flag

6. ai-native userIssue scoped/least-privilege API credentials for an agent

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed8/10

   Stripe documents Restricted API Keys (RAKs) as 'the recommended type of key for API keys you give to AI agents' with granular permission scoping, plus scoped Shared Payment Tokens with currency/amount/expiration limits and revocation, and MCP server actions gated by human approval tokens — together giving strong least-privilege credentialing for agents. missing for 10: independent/hands-on verification of RAK permission granularity in practice, and no evidence of automated credential rotation or audit-logging specifics for agent-issued keys.

   • [claimed-docs] “RAKs are the recommended type of key for API keys you give to AI agents.”
   • [claimed-docs] “Stripe recommends always using RAKs instead of unrestricted secret keys, especially when giving a key to an AI agent. Use RAK permissions to…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “Coding agents should install the Stripe CLI (npm i -g @stripe/cli) and run the command stripe sandbox create --help to provision an anonymou…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed4/10

   There's no explicit 'issue a scoped API key/token for an agent' feature for the general Agent Toolkit or REST API surface, but agent-specific payment tokens are documented as scoped/least-privilege by design — the ACP delegated token is 'bound to your merchant ID and includes amount and time restrictions that you can configure,' and the MCP server enforces OAuth with PKCE and dynamic client registration rather than static keys. This shows some least-privilege credentialing pattern for agent checkout flows but not a general-purpose scoped-credential system for arbitrary agent API access. Missing for 10: documented scope/permission tiers (e.g., read-only vs write) for the Agent Toolkit or REST OAuth tokens, and any self-serve UI/API to mint restricted credentials per agent.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”

   proof ↗⚑ flag

7. ai-native userBuild against official SDKs

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed9/10

   Stripe ships extensive official SDK/CLI/protocol tooling for AI-native builders: a hosted MCP server (docs + runtime probe confirming live OAuth-gated endpoint), Link CLI published on npm with runtime verification, agent skills catalog served at a well-known JSON endpoint, restricted API keys guidance for agents, MPP/x402 machine-payment protocols with open spec and mppx conformance CLI, and one-prompt agent build flows. This is corroborated by independent runtime probes (npm resolution, live endpoint checks), not just vendor docs. missing for 10: no independent third-party developer testimonial/case study of building against these SDKs (only Stripe-run runtime probes and docs), and ACP/MCP write actions still require human approval gating rather than fully autonomous agent operation.

   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API”
   • [claimed-docs] “Install instructions that help AI agents create more accurate Stripe integrations.”
   • [claimed-docs] “RAKs are the recommended type of key for API keys you give to AI agents.”
   • [claimed-docs] “Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to comple…”
   • [github] “The CLI can produce one of three credential types: A virtual card (PAN)... A Link Pay Token (LPT)... A Shared Payment Token (SPT)”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [probe] “PROBE runtime (recorded 2026-09-14): the machine-readable skills catalog is live at the documented keyless well-known URL — https://docs.str…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullprobed8/10

   PayPal publishes an official Agent Toolkit SDK (npm/PyPI, versioned and actively published) integrating with MCP, OpenAI Agents SDK, LangChain, CrewAI, Vercel AI SDK, and Bedrock, plus documented pre-built functions for orders, invoices, subscriptions, refunds, disputes, and catalog management, backed by REST API docs, sandbox testing, and webhooks. This is corroborated by runtime probes confirming the toolkit is live on public registries and the MCP server is reachable. Missing for 10: independent third-party developer reviews/case studies of building against these SDKs beyond PayPal's own docs and registry probes.

   • [claimed-docs] “PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using Amazon Bedrock, CrewAI, LangChain, Model Contex…”
   • [claimed-docs] “PayPal built an MCP server that lets merchants use natural language with their favorite MCP clients. This helps users complete business task…”
   • [claimed-docs] “Integrate with PayPal APIs to access orders, invoices, subscriptions, shipment tracking, transaction details, and dispute management through…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [github] “create_order: Create an order in PayPal system based on provided details”

   proof ↗⚑ flag

8. ai-native userSubscribe to events via webhooks

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed8/10

   Multiple agentic-commerce flows emit dedicated webhooks that a developer/agent-builder can subscribe to: order completion (`checkout.session.completed` tagged with agent), catalog import status (`v2.commerce.product_catalog.imports.*`), OCA embedded checkout completion (`delegated_checkout.requested_session.completed`), and SPT state transitions (`shared_payment.issued_token.used`, etc.), all part of Stripe's standard event/webhook infrastructure applied to agentic commerce. Missing for 10: explicit doc snippet on creating/registering a webhook endpoint for these events and independent hands-on confirmation of subscription setup.

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullclaimed7/10

   PayPal's webhook API is explicitly documented: merchants can subscribe up to 10 webhook URLs per app, scoped to specific event types, with durable retry (25 attempts over 3 days) and signature verification guidance — enabling event-driven, agentic integrations. Missing for 10: explicit agentic-commerce-specific event types (e.g., cart/order events tied to AI checkout flows) and independent/hands-on corroboration beyond first-party docs.

   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”

   proof ↗⚑ flag

Agentic features

1. ai-native userGet AI-generated insights and suggestions from my data inside the product

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed4/10

   Evidence shows narrow instances of AI-generated insight/suggestion from user data — buyers get 'personalized recommendations' during chat commerce, and the Link CLI wallet lets an agent 'retrieve permissioned financial data to answer questions about spending, balances, and trends' (e.g., 'What did I spend on dining this month?'). This is real but thin — it's a side feature of a payments/commerce infra product, not a dedicated insights/analytics capability with elaboration or hands-on validation. Missing for 10: dedicated in-product analytics/insights surface beyond brief doc mentions, detail on recommendation quality/methodology, and independent corroboration that these insight features work as described.

   • [claimed-docs] “Buyers can browse products, get personalized recommendations, and complete purchases without leaving the conversation.”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [claimed-docs] “retrieve permissioned financial data to answer questions about spending, balances, and trends.”
   • [claimed-docs] ““What did I spend on dining this month?””

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence documents PayPal's agentic commerce stack (Agent Ready, Store Sync, cart APIs, agent toolkit, MCP server) purely as infrastructure enabling AI agents to browse catalogs, create carts, and complete transactions on a merchant's behalf — none of it shows PayPal analyzing a user's/merchant's own data to surface AI-generated insights or suggestions inside the product (e.g., analytics dashboards, business recommendations). The dispute/order/refund tools are action endpoints for agents to execute tasks, not insight-generation features.

   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “PayPal built an MCP server that lets merchants use natural language with their favorite MCP clients. This helps users complete business task…”

   proof ↗⚑ flag

2. ai-native userSet up automations that run autonomously in the background

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Machine payments (x402, MPP) let agents transact programmatically without a human in the loop for each call, and SPTs allow scoped, usage-limited autonomous spending — supporting some background automation. However, the same evidence pack documents pervasive human-approval gates for higher-value or write actions ('Stripe requires human confirmation before it takes certain stripe_api_write actions' and 'Customers approve each request on the Link website or in the mobile app'), which directly limits true unattended background autonomy for most commerce flows. Missing for 10: a documented persistent/scheduled automation or workflow feature that runs entirely without any human approval step across the product's full scope, and independent evidence of such automations running successfully unattended.

   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   PayPal provides the building blocks for background automation — an agent toolkit with pre-built functions (orders, refunds, disputes, subscriptions) usable in LangChain/CrewAI/Bedrock workflows, an MCP server for natural-language triggered actions, and webhooks that fire asynchronously with retry logic — enabling autonomous background agent workflows. However, PayPal's own docs explicitly recommend keeping 'human oversight in areas where personal judgment matters most,' and much of the deeper commerce automation (Store Sync, agentic commerce services) is gated behind manual application forms rather than self-serve autonomous setup. Missing for 10: evidence of a fully autonomous, unattended scheduling/orchestration layer, independent confirmation of agents running unattended in production, and self-serve (non-form-gated) enablement.

   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”

   proof ↗⚑ flag

3. ai-native userOperate the product with natural-language commands

   weight 2 · round drawn
   Stripe Agentic Commerce~partialprobed6/10

   Stripe's agentic-commerce stack is built around natural-language-driven flows: buyers can browse, get recommendations, and 'complete purchases without leaving the conversation,' Link CLI supports natural queries like 'Buy me this book from Stripe Press' and 'What did I spend on dining this month?', and the MCP server/skills let AI agents interact with the API conversationally. However, this is fundamentally an API/protocol/SDK product for developers and agent-builders, not an end-user interface itself — the 'natural language' surface is provided by the calling LLM/agent (ChatGPT, etc.), and most interactions (SPT issuance, key management, catalog feeds, checkout hooks) are structured API/CLI calls rather than NL commands to Stripe itself. Missing for 10: evidence of Stripe's own product surface directly accepting free-form natural-language commands (as opposed to relying on a third-party agent/LLM to translate NL into API calls), and independent hands-on confirmation that the documented NL examples work end-to-end in production rather than as illustrative docs snippets.

   • [claimed-docs] “Buyers can browse products, get personalized recommendations, and complete purchases without leaving the conversation.”
   • [claimed-docs] ““Buy me this book from Stripe Press””
   • [claimed-docs] ““What did I spend on dining this month?””
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API”
   • [claimed-docs] “Install instructions that help AI agents create more accurate Stripe integrations.”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   PayPal supports natural-language operation via its MCP server and agent toolkit (LangChain, OpenAI Agents SDK, CrewAI, etc.), letting AI agents issue commands for orders, refunds, disputes, and shopping carts in conversational form, and a runtime probe confirms the MCP endpoint is live. However, the 'user' operating this isn't typically an end consumer speaking to PayPal directly — most flows are merchant-hosted or gated behind Braintree/Google/form approval, and the natural-language surface is mediated through third-party chat assistants (ChatGPT, Gemini) or developer-built agents rather than a native PayPal NL interface. missing for 10: independent/hands-on evidence of an end user directly issuing free-form natural-language commands to PayPal itself (not via a third-party chat assistant or developer-built agent), and confirmation the MCP server/agent toolkit works smoothly without heavy setup/gating.

   • [claimed-docs] “PayPal built an MCP server that lets merchants use natural language with their favorite MCP clients. This helps users complete business task…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”

   proof ↗⚑ flag

Api quality

1. ai-native userExplore an interactive API reference with runnable examples

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialprobed4/10

   Evidence shows agent-facing tooling that lets an AI actually exercise the API — a sandbox CLI that provisions working keys ("stripe sandbox create --help"), an MCP server exposing callable API tools, and a machine-readable skills/llms.txt catalog — but nothing in the pack describes a classic interactive API reference page with inline runnable code examples; the docs are plain markdown (per the .md probes) and the openapi.json/swagger.json spec paths all 404. missing for 10: a documented interactive reference UI (e.g. try-it/run-in-browser code samples), an OpenAPI/Swagger spec for the agentic-commerce endpoints, independent confirmation that examples execute against live data.

   • [claimed-docs] “Coding agents should install the Stripe CLI (npm i -g @stripe/cli) and run the command stripe sandbox create --help to provision an anonymou…”
   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API and search Stripe’s kno…”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the machine-readable skills catalog is live at the documented keyless well-known URL — https://docs.str…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/agentic-commerce.md # Agentic commerce Sell through agents, or embed commerce into your …”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence shows static API reference pages (createcart, getcart) and REST API docs, plus a sandbox environment for testing, but nothing describing an interactive reference with embedded runnable/try-it-out examples (e.g., live code execution in docs, Swagger-style console). Missing for 10: any mention of an interactive API console, embedded runnable code snippets, or in-docs request execution.

   • [claimed-docs] “Creates a new cart with the provided PayPalCart object. This is called by PayPal Shopping Cart when an AI agent or customer initiates a chec…”
   • [claimed-docs] “Retrieves the current state of a cart. Used by PayPal Shopping Cart to check cart status, validation issues, and current totals.”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”

   proof ↗⚑ flag

2. ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)

   weight 2 · round drawn
   Stripe Agentic Commerce—none0/10

   The evidence pack shows a direct probe for an OpenAPI/Swagger spec at all standard locations (openapi.json, swagger.json, etc.) returning 404 across the board, and no other citation surfaces a downloadable machine-readable API spec for the Agentic Commerce endpoints. While Stripe ships llms.txt, a skills.json index, and an MCP server, none of these constitute a downloadable OpenAPI/equivalent spec for the story as written.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack documents REST API references, SDKs, webhooks, and endpoint descriptions (e.g., createcart/getcart docs), but nowhere mentions a downloadable OpenAPI/Swagger file or other machine-readable spec artifact for the agentic-commerce or general PayPal APIs. This axis is clearly applicable to an API-based commerce product, but no evidence shows PayPal exposing such a spec for download.

   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”

   proof ↗⚑ flag

3. ai-native userTest against a sandbox environment without touching production data

   weight 1 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed7/10

   Stripe explicitly documents a sandbox provisioning flow for coding agents: 'run stripe sandbox create --help to provision an anonymous Stripe sandbox with working API keys. No account registration required,' which directly satisfies testing without touching production data. Missing for 10: independent/hands-on confirmation that agentic-commerce-specific flows (ACS, SPTs, machine payments) work in this sandbox, and documentation on how test-mode agent transactions differ from production before going live.

   • [claimed-docs] “Coding agents should install the Stripe CLI (npm i -g @stripe/cli) and run the command stripe sandbox create --help to provision an anonymou…”

   proof ↗⚑ flag

   PayPal Agentic Commerce✓fullclaimed8/10

   PayPal documents a dedicated sandbox that mirrors production, allowing developers/agents to test agent-toolkit tools, cart API, and checkout flows without touching live accounts, and confirms the same REST rails (Orders, cart API) run against sandbox credentials. Missing for 10: no independent/hands-on verification of sandbox parity for agentic-commerce-specific flows (cart API, ACP/UCP checkout) and no explicit documentation of sandbox mode for the MCP server or agent toolkit specifically.

   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”

   proof ↗⚑ flag

4. ai-native userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Stripe Agentic Commerce—none0/10

   The evidence pack covers extensive agentic-commerce features (SPTs, MCP, MPP, Link CLI, ACP) but contains no documentation of API version strings, an upgrade/versioning scheme, or a deprecation policy for these agentic-commerce endpoints — several features are explicitly flagged as 'private preview' or waitlisted rather than versioned/stable, but no deprecation policy is cited anywhere.

   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack documents PayPal's agentic-commerce APIs (createcart, getcart, agent toolkit, REST APIs) including a versioned path like /v1/createcart, but nowhere is there any documented deprecation policy, versioning strategy, or backward-compatibility commitment for AI agents to rely on. The axis is clearly applicable (PayPal is an API-first commerce platform) but no supporting evidence exists.

   • [claimed-docs] “Creates a new cart with the provided PayPalCart object. This is called by PayPal Shopping Cart when an AI agent or customer initiates a chec…”
   • [claimed-docs] “Retrieves the current state of a cart. Used by PayPal Shopping Cart to check cart status, validation issues, and current totals.”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”

   proof ↗⚑ flag

1. ai-native userPerform bulk operations across many items at once

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Stripe provides some bulk-like primitives — sellers can upload a whole product catalog via CSV feed (up to 4GB) and 'bulk fulfill orders' via the List CheckoutSessions endpoint — but these are seller-side backend operations, not an AI-native user-facing bulk operation across many purchases/items in one agent action. There's no evidence of an agent being able to, e.g., issue one command to buy/refund/manage many items or transactions at once as an end-user capability. missing for 10: agent-facing bulk purchase/checkout across many items in one call, bulk SPT/token operations, batch API endpoints exposed to AI agents beyond seller catalog upload and fulfillment listing, independent hands-on confirmation of bulk agent workflows.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence shows only single-item agent tools (create_order, create_refund, get_dispute, etc.) and a one-time bulk catalog feed upload for merchant setup, but no documented bulk/batch operation capability (e.g., batch refunds, batch order processing, multi-item transaction endpoints) that an AI agent could invoke across many items at once.

   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”

   proof ↗⚑ flag

2. ai-native userDefine rules that trigger actions automatically on events

   weight 3 · round drawn
   Stripe Agentic Commerce—none0/10

   Evidence shows Stripe emits webhooks (checkout.session.completed, SPT state-transition events, import success/failure events) that developers can build on, but there is no documented feature letting an AI-native user define custom rules/conditions that automatically trigger actions on events — in fact, agent write-actions via MCP explicitly require human approval before execution, the opposite of automatic rule-triggered action.

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence shows PayPal offers webhooks (event notifications) and an agent toolkit with individual API tools, but nothing resembling a rules engine where a user defines 'if event X then action Y' automation — webhooks merely deliver raw event notifications that require the merchant to write custom code to act on, not a rule-definition interface. Missing for 10: any rule-builder/automation-trigger UI or API, conditional logic definition, and evidence of automatic action execution tied to rules rather than manual webhook handling.

   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”

   proof ↗⚑ flag

Fees transparency

1. merchantThe fees for agent-originated transactions are published — I can price agentic channels without asking sales

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullcommunity7/10

   Stripe explicitly publishes the flat card-processing rates (2.9%+30¢ domestic, plus documented adders for manual entry/international/currency conversion) and states these standard rates are what agent-originated orders inherit, with 'no separate agentic-channel fee schedule disclosed' — meaning a merchant can price ACS-driven sales from the public pricing page without a sales call. Missing for 10: no dedicated line-item fee schedule specifically labeled for agent/ACS transactions (only an inference that standard rates apply), and no clarity on whether third-party platform fees (e.g., ChatGPT's own merchant commission referenced in community discussion) stack on top for ACP-routed orders.

   • [claimed-docs] “stripe.com/pricing publishes the flat rates that agent-driven orders inherit, with no separate agentic-channel fee schedule disclosed: Stand…”
   • [community] “"Merchants pay a small fee on completed purchases, but the service is free for users, doesn't affect their prices, and doesn't influence Cha…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed5/10

   Evidence shows agent-originated transactions run through existing Braintree/PayPal Checkout rails and thus inherit the standard published US commercial fee schedule (2.99–4.99% + fixed fee), so merchants aren't quoted a hidden or sales-negotiated rate for the payment processing itself. However, the same evidence explicitly states 'No separate agentic/AI-channel fee schedule is published,' and access to Agent Ready/Store Sync/ACS is gated behind a sales-contact form, leaving open whether any additional platform or enablement fee applies to the agentic channel — so a merchant cannot fully price the agentic channel without engaging sales for onboarding. Missing for 10: an explicit agentic-channel fee page or fee line item, and self-serve (non-form-gated) enrollment that would let a merchant confirm total costs without sales contact.

   • [claimed-docs] “The published US commercial rates that agent-initiated orders inherit: "PayPal Checkout" and "PayPal Guest Checkout" at "3.49% +" fixed fee,…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”

   proof ↗⚑ flag

Payouts

1. merchantAgent-originated payments settle through documented payout and reporting flows, reconcilable like any other channel

   weight 1 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed8/10

   Agent-originated orders flow through standard Stripe primitives - checkout.session.completed events, Dashboard transaction tagging/filtering by agent, an agent_details field, bulk fulfillment, and standard Refunds API - with pricing docs confirming no separate agentic fee schedule and machine payments explicitly stating settlement/reporting 'work the same as any other payment in Stripe'. missing for 10: independent/third-party confirmation that reconciliation works in practice, and removal of the 'Private preview' gate on agent_details

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [claimed-docs] “stripe.com/pricing publishes the flat rates that agent-driven orders inherit, with no separate agentic-channel fee schedule disclosed: Stand…”
   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed6/10

   Agent-originated orders flow into the merchant's existing order management, payment, refund/dispute, and webhook systems (create_order, refund/dispute tools, webhooks with signature verification, existing Orders v2/Braintree rails), and docs explicitly state Store Sync orders arrive 'just as it would from any other channel,' suggesting standard reconciliation. However, there is no dedicated agentic-channel reporting/settlement documentation, no separate transaction tagging or reconciliation report distinguishing agent-originated payments, and fee schedules show no distinct line item for agent transactions, making channel-level reconciliation an inference rather than a documented flow. missing for 10: dedicated agent-channel reporting/settlement statements, explicit reconciliation tooling or transaction tags identifying agent-originated payments, independent merchant confirmation of clean reconciliation in production.

   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [claimed-docs] “The published US commercial rates that agent-initiated orders inherit: "PayPal Checkout" and "PayPal Guest Checkout" at "3.49% +" fixed fee,…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”

   proof ↗⚑ flag

Attribution analytics

1. merchantOrders are attributed to the originating agent and I can report on or filter transactions by agent

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullclaimed8/10

   Stripe's seller docs explicitly state that agent-driven orders generate a unique checkout.session.completed event, appear on the Dashboard Transactions page 'tagged with the originating agent,' and merchants 'can filter transactions by agent names,' with a documented agent_details field and bulk-fulfillment support via the CheckoutSessions API. Missing for 10: independent/hands-on confirmation of the attribution UI in practice, and the agent_details field is noted as Private preview rather than GA.

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence covers agent-initiated checkout flows (ACP/UCP/Store Sync/cart API/agent toolkit) but nowhere describes orders being tagged with an originating-agent identifier, nor any merchant dashboard/report/filter capability by agent; in fact Store Sync docs state orders arrive 'just as it would from any other channel,' suggesting no special agent attribution.

   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”

   proof ↗⚑ flag

Onboarding to agents

1. merchantI have a documented self-serve path to start accepting agent traffic — enable a setting, sign an agreement, or publish an endpoint

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed7/10

   Stripe documents a genuine self-serve merchant path: onboard as a seller in the Dashboard, review and accept per-agent terms, publish a CSV product catalog via presigned-URL API, and receive human-approval-gated agent orders tagged in the Transactions page. However this is explicitly gated - Agentic Commerce Suite is private preview and restricted to US, Canada, and select European countries, and the custom ACP-style integration explicitly requires joining a waitlist rather than pure self-serve enablement. Missing for 10: evidence of unrestricted GA availability outside private preview, and a self-serve (non-waitlist) path for the custom checkout-endpoint integration.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”
   • [claimed-docs] “Stripe's agentic-commerce hub is explicitly chat-native: "Buyers can browse products, get personalized recommendations, and complete purchas…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed5/10

   There is documented onboarding material (Agent Ready guides, Store Sync docs, catalog feed specs), but the path is explicitly gated rather than self-serve: PayPal's own docs state that agentic commerce services access is 'form-gated, not self-serve' requiring a form submission and manual follow-up from PayPal's AI team, Store Sync eligibility requires an existing Braintree/Orders v2 account plus the same approval gate, and the UCP/Google Pay path requires separate 'merchant registration with Google. Contact Google for enablement.' Agent Ready itself assumes an existing Braintree merchant relationship rather than a simple toggle or self-issued agreement. Missing for 10: a true no-approval self-serve enablement (setting flip, standard clickthrough agreement, or endpoint publish without a contact-form/manual review step), and confirmation of approval turnaround time.

   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Catalog publishing is feed-file based: "PayPal supports 3 feed specifications", including the Google Product Feed ("you can reuse it to powe…”

   proof ↗⚑ flag

Traffic controls

1. merchantRate-limit, allow, or deny specific agents and agent platforms accessing my catalog and checkout

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Merchants get agent-level allow/deny control via OCA consent ("Find the agent you want to sell through and review its terms", approval requests, and either party can terminate an OCA at any time) plus per-agent order attribution/filtering in the Dashboard, but there is no documented rate-limiting mechanism for specific agents or platforms. missing for 10: explicit per-agent or per-platform rate-limit/throttling controls, granular allow/deny at the platform (vs individual agent) level, and any dashboard UI for setting quotas.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence shows PayPal gates agentic-commerce access at the platform/onboarding level (Braintree account, Google UCP registration, PayPal AI-team approval form) and gives merchants brand/communication control, but nothing describes merchant-configurable rate limits or per-agent/per-platform allow/deny controls over catalog or checkout access.

   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”

   proof ↗⚑ flag

1. ai-native userDo everything through the API that I can do in the UI

   weight 2 · round drawn
   Stripe Agentic Commerce~partialclaimed6/10

   Stripe's docs show an extensive API surface for agentic commerce (Product Catalog Import API, Checkout Sessions, Shared Payment Tokens, MCP server exposing Stripe API tools, restricted API keys, CLI/sandbox tooling) letting agents perform most seller/agent/consumer actions programmatically, with the MCP server explicitly wrapping the API. However evidence also documents real UI/human dependencies: seller onboarding is described as 'Dashboard self-serve', and Stripe's own MCP server 'requires human confirmation' via a Dashboard approval URL before certain writes (refunds, outbound payments), meaning not all actions can be completed purely via API without a human UI step. Missing for 10: evidence that seller onboarding and gated write actions (refunds, payouts) can be fully completed via API without a Dashboard/human approval step, and independent confirmation of full UI/API feature parity.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “The Stripe Model Context Protocol (MCP) server provides tools that AI agents can use to interact with the Stripe API”
   • [claimed-docs] “RAKs are the recommended type of key for API keys you give to AI agents.”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed6/10

   PayPal's agent toolkit and REST APIs expose a broad set of merchant operations (orders, invoices, subscriptions, refunds, disputes, shipment tracking, catalog management) that mirror much of what's doable in the PayPal/Braintree dashboards, and there's a documented MCP server, SDKs, and sandbox for building against these. However, the flagship agentic-commerce paths are heavily gated rather than self-serve: Agent Ready requires a Braintree account, Store Sync and Agentic Commerce Services require a manual form-based approval process, UCP requires separate Google merchant registration, and Store Sync itself is scoped only to US physical goods in USD — none of which are pure API-driven equivalents of UI actions. Missing for 10: explicit UI-vs-API feature-parity mapping, self-serve (non-gated) access to the agentic commerce APIs, and confirmation that all dashboard-level configuration/reporting is API-accessible without approval gates.

   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “The PayPal sandbox "is a self-contained, virtual testing environment that simulates the live PayPal production environment" — "a shielded sp…”

   proof ↗⚑ flag

2. ai-native userRead the product's source under an open license

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialprobed4/10

   The core Agentic Commerce Suite (checkout backend, SPT infrastructure, MCP server) is proprietary Stripe infrastructure with no published source, but the Agentic Commerce Protocol is explicitly 'open source and community-designed under the Apache 2.0 license,' and the Link CLI client has a public GitHub repo (stripe/link-cli). This gives an AI-native user real, licensed source to read for parts of the ecosystem, but not for the product's core commerce/payment engine. Missing for 10: open-licensed source for the actual ACS backend/checkout service, MCP server implementation, and MPP/x402 server-side code — only the protocol spec and a thin CLI client are open.

   • [claimed-docs] “"ACP is open source and community-designed under the Apache 2.0 license. Businesses can implement the specification to transact with any AI …”
   • [github] “The CLI can produce one of three credential types: A virtual card (PAN)... A Link Pay Token (LPT)... A Shared Payment Token (SPT)”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   PayPal's agentic commerce services (Store Sync, ACP/UCP integrations, agent toolkit) are proprietary hosted APIs and closed-source integrations; the agent-toolkit SDK is published on npm/PyPI but no evidence indicates an open-source license for its source, and core commerce services (Store Sync, cart API, MCP server backend) are entirely closed, gated behind forms and merchant accounts. No evidence pack item shows an open-license repository for the product's source.

   ⚑ flag

Order events

1. developerMy agent or backend can subscribe to order lifecycle events — created, updated, fulfilled, canceled — pushed programmatically

   weight 3 · round drawn
   Stripe Agentic Commerce~partialclaimed5/10

   Stripe documents at least one lifecycle push event for agent orders — checkout.session.completed fires when an agent completes an order, tagged with agent attribution and filterable in the Dashboard — plus a delegated_checkout.requested_session.completed webhook for embedded checkout completion, and separate webhooks exist for catalog-import and SPT-token state changes. However, there is no documented event set covering the full requested lifecycle (created, updated, fulfilled, canceled) for orders themselves; canceled/fulfilled-specific webhooks aren't shown, and bulk fulfillment is presented as a pull (List CheckoutSessions) rather than a push mechanism. Missing for 10: explicit order.created/updated/canceled webhook events, a documented order-events reference, and independent/hands-on confirmation of subscription reliability.

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed5/10

   PayPal's generic webhooks system (developer.paypal.com/api/webhooks/overview) lets a merchant subscribe up to 10 URLs to specific event types, with durable retry (up to 25 attempts/3 days) and signature verification — and the agent-toolkit/cart API run on the same REST/order rails, so this infrastructure would carry order events. However, no evidence pack item explicitly names order-lifecycle event types (ORDER.CREATED/UPDATED/CANCELLED/fulfilled) tied to the agentic-commerce cart or Store Sync flows — those docs describe merchant-hosted reverse endpoints (createcart/getcart) rather than push-based lifecycle events for agents. Missing for 10: explicit order-lifecycle event-type list for agentic commerce, confirmation that Store Sync/cart orders emit standard webhook events, and any hands-on/independent proof of an agent successfully subscribing to these events.

   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The companion read endpoint keeps agent-side cart state fresh: get cart "Retrieves the current state of a cart. Used by PayPal Shopping Cart…”

   proof ↗⚑ flag

Post purchase status

1. ai-native userMy agent can query the status and tracking of an order it placed, after the purchase

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce—none0/10

   Evidence covers checkout completion, order attribution in the Stripe Dashboard, and bulk fulfillment tools for sellers, but nothing shows an agent being able to query order status or shipping/tracking information after a purchase is placed — no order-status API, webhook, or tool is documented for the agent side post-purchase.

   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   PayPal's agent toolkit and REST surface explicitly advertise agent access to 'orders...shipment tracking, transaction details' for post-purchase queries, and the agent-tools page lists concrete order-related functions (create_order, pay_order, create_shipment) alongside webhooks for order events. However, no specific 'get_order_status' or 'track_shipment' tool call is named or demonstrated end-to-end, so the tracking-query capability is asserted at a high level rather than shown working. Missing for 10: an explicitly named order-status/tracking query tool/function, and independent or hands-on confirmation that an agent can actually retrieve tracking info post-purchase.

   • [claimed-docs] “Integrate with PayPal APIs to access orders, invoices, subscriptions, shipment tracking, transaction details, and dispute management through…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”

   proof ↗⚑ flag

Delegated credentials

1. ai-native userHand my agent a scoped, limited-use payment credential — amount limits, expiry, merchant restrictions — instead of a raw card number

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Stripe ships exactly this: Shared Payment Tokens grant sellers scoped access with explicit usage_limits (currency, max_amount, expires_at), merchant/seller scoping, and revocation, and Link CLI issues one-time-use virtual cards/LPTs/SPTs to agents rather than raw PANs, with human approval gates on spend. This is corroborated by docs, GitHub source, and live npm/runtime probes. missing for 10: independent hands-on third-party audit of the credential-scoping mechanics (only vendor docs + probes, no external security review), and broader geographic/GA availability (features are private preview / US-Canada-EU only).

   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [github] “The CLI can produce one of three credential types: A virtual card (PAN)... A Link Pay Token (LPT)... A Shared Payment Token (SPT)”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [claimed-docs] “Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to comple…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed6/10

   PayPal's ACP/UCP flows use Braintree single-use payment tokens/nonces that are explicitly described as 'bound to your merchant ID and includes amount and time restrictions that you can configure' — a genuine scoped, limited-use credential rather than a raw card number, and Google's UCP path similarly returns a tokenized, single-use credential. However this is merchant-side infrastructure (tokens generated by the merchant's Braintree integration for a specific checkout), not a consumer/end-user-controlled credential that an AI-native user personally hands to their own agent with self-set amount/expiry/merchant limits; there's no evidence of a user-facing wallet-style scoped credential issuance flow. missing for 10: end-user-facing UI/API for a consumer to mint their own scoped credential (set their own amount cap, expiry, merchant allowlist) independent of a specific merchant checkout session, and independent/hands-on confirmation of these restriction configs working in practice.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”

   proof ↗⚑ flag

Machine payments

1. developerMonetize an API or service with HTTP 402 machine payments that agents pay per call without a checkout UI

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed9/10

   Stripe explicitly documents HTTP 402 machine payments (x402 and MPP) letting agents pay per API call with no checkout UI, backed by a runtime probe confirming the MPP spec and mppx SDK are live, plus a conformance CLI (mppx validate) and one-prompt build path. missing for 10: independent/hands-on developer report of a working 402 integration beyond Stripe's own docs and probes.

   • [claimed-docs] “Machine payments let agents pay for APIs and services programmatically.”
   • [claimed-docs] “x402 is a protocol for internet payments... your server returns an HTTP 402 response with payment details, including a Stripe deposit addres…”
   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   All evidence describes PayPal's agentic commerce as checkout-flow based (carts, ACP/UCP delegated tokens, Store Sync, agent toolkit for orders/refunds/disputes) — none of it describes HTTP 402 status-code machine payments or per-call API monetization without a checkout UI, which is a distinct technical pattern. Since PayPal is a payments platform, this axis could plausibly apply, but there is no evidence it ships or supports the 402 pay-per-call mechanism described in the story.

   ⚑ flag

Settlement rails

1. developerSettle agent-initiated payments in stablecoins over documented rails

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialprobed6/10

   Stripe explicitly documents stablecoin-denominated agent payments over documented rails (MPP on Tempo/USDC.e and Solana/USDC, x402 on Base/USDC) with published minimums (0.01 USDC) and geographic scope, and a runtime probe confirms the MPP spec and mppx SDK are live. However, the same docs state payments 'settle in fiat' into the Stripe balance rather than being settled to the developer in stablecoins, so the 'settle in stablecoins' half of the story is not actually delivered — it's stablecoin-denominated payment collection with fiat settlement. Missing for 10: actual stablecoin-denominated settlement (not fiat conversion), independent/hands-on developer corroboration beyond Stripe's own docs.

   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [claimed-docs] “x402 is a protocol for internet payments... your server returns an HTTP 402 response with payment details, including a Stripe deposit addres…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [claimed-docs] “You can build an API that uses MPP with a single prompt to your coding agent”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack details PayPal's agentic commerce rails extensively (ACP/UCP via Braintree tokens, agent toolkit, cart APIs, webhooks, fees) but none of it mentions stablecoins, PYUSD, crypto, or blockchain-based settlement rails for agent-initiated payments — all documented settlement is via traditional card/Braintree processing and standard PayPal fee schedules.

   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”
   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “The published US commercial rates that agent-initiated orders inherit: "PayPal Checkout" and "PayPal Guest Checkout" at "3.49% +" fixed fee,…”

   proof ↗⚑ flag

Wallets

1. ai-native userProvision a wallet my agent holds and spends from — funded and controlled by me, with balance visibility

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Link CLI documents a customer-controlled wallet from which the agent retrieves one-time-use payment credentials, with balance/transaction visibility ('reads transaction history, balances, and account details') and human approval on every spend via the Link app/website, and is confirmed live on npm by runtime probe. missing for 10: independent/hands-on user reports of actually funding and using the wallet end-to-end (only vendor docs and a package-registry probe, no third-party usage account).

   • [claimed-docs] “Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to comple…”
   • [claimed-docs] “Your agent reads transaction history, balances, and account details from connected bank accounts and credit cards.”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [github] “The CLI can produce one of three credential types: A virtual card (PAN)... A Link Pay Token (LPT)... A Shared Payment Token (SPT)”
   • [github] “A virtual card (PAN) for use with a standard web checkout form. The issued card works anywhere, and is not restricted to Link-enabled seller…”
   • [claimed-docs] “Customers approve each request on the Link website or in the mobile app.”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   All evidence describes merchant-side infrastructure (Agent Ready, Store Sync, agent toolkit, cart APIs) that lets AI agents check out with merchants or lets merchants build agents for order/refund/dispute management — none of it describes a consumer-provisioned wallet that a user funds, an agent holds/spends from, with balance visibility. The delegated tokens in ACP/UCP are single-use, merchant-bound checkout credentials, not a persistent user wallet.

   ⚑ flag

1. ai-native userControl data retention and deletion

   weight 2 · round drawn
   Stripe Agentic Commerce—none0/10

   The evidence pack covers token/authorization revocation (SPT revocation, OCA termination) and human-approval gates for payment actions, but nothing about user-controlled data retention policies, data export, or deletion of stored personal/transaction data. Privacy-posture around retention/deletion is a fair question for a payments platform handling PII and financial data, but no documentation addresses it.

   ⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack covers agentic commerce protocols, cart APIs, webhooks, fees, and toolkit integrations, but contains no mention of data retention policies, deletion controls, or privacy settings that an AI-native user could invoke to manage stored data. This axis plausibly applies to a payments platform handling AI agent transaction data, but no supporting evidence exists.

   ⚑ flag

Ecosystem adoption

1. developerAt least two independent parties beyond the vendor implement the protocol or platform — it is an ecosystem, not a private integration

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialprobed6/10

   Stripe documents ACP as "open source and community-designed under the Apache 2.0 license" co-developed with OpenAI (who implemented it in ChatGPT as "the first AI platform"), and MPP as an open protocol "co-authored by Stripe and Tempo," with x402/network-token integrations naming Mastercard's Agent Pay and Visa's Intelligent Commerce as third-party implementers. Independent runtime probes confirm mpp.dev is a separate, non-Stripe-branded spec site with governance and SDK on public npm, supporting genuine multi-party involvement rather than a private integration. Missing for 10: independent (non-Stripe-authored) case studies or hands-on reports of a third seller/agent platform beyond OpenAI/Tempo actually running the protocol in production, and clearer separation of protocol governance from Stripe's own docs.

   • [claimed-docs] “"ACP is open source and community-designed under the Apache 2.0 license. Businesses can implement the specification to transact with any AI …”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [claimed-docs] “The ACP-style custom integration has sellers "implement a reverse API that defines the requests Stripe sends to your commerce backend and th…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence shows PayPal/Braintree acting as an implementer of protocols defined by other parties (OpenAI's ACP, Google's UCP, and the Anthropic-originated MCP), but nothing shows independent third parties implementing PayPal's own agentic-commerce platform (Cart API, Store Sync, agent toolkit) as an open ecosystem. Merchants and partners (Wix, Shopware, BigCommerce) are consumers/integrators of PayPal's platform, not independent implementers of its protocol.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Store Sync is the full-catalog path, "enabling AI agents to discover your products, create and manage shopping carts, and complete purchases…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”

   proof ↗⚑ flag

Spec publication

1. developerThe commerce protocol this product speaks is published as an open, versioned specification I can read and implement without signing anything

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Stripe explicitly documents that the Agentic Commerce Protocol (ACP) it speaks is "open source and community-designed under the Apache 2.0 license" and that "businesses can implement the specification to transact with any AI agent or payment processor" with no signup required to read or implement it — only optional ChatGPT marketplace participation needs an application, which is a separate distribution channel, not the spec itself. This is corroborated by similar open-protocol framing for MPP ("an open protocol... co-authored by Stripe and Tempo", live spec at mpp.dev) and x402, and a runtime probe confirms the MPP spec is publicly fetchable without auth. Missing for 10: explicit version-numbering/changelog evidence for the spec, and independent (non-Stripe) confirmation that implementers never need a signed agreement to build against the published spec.

   • [claimed-docs] “"ACP is open source and community-designed under the Apache 2.0 license. Businesses can implement the specification to transact with any AI …”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Machine Payments Protocol is published in the open — https://mpp.dev/ answers a plain GET with the …”
   • [claimed-docs] “MPP "is an open protocol that lets agents pay for your APIs and services programmatically without a checkout UI, co-authored by Stripe and T…”
   • [claimed-docs] “x402 is a protocol for internet payments... your server returns an HTTP 402 response with payment details, including a Stripe deposit addres…”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   The evidence pack shows PayPal implements OpenAI's ACP and Google's UCP protocols and tells developers to 'follow the ACP agentic checkout specification,' but never provides or links a version number, license, or spec document for ACP/UCP that a developer could read independently of PayPal. Instead, everything documented is gated: Braintree merchant account required, PayPal AI team form-approval required for agentic commerce services, and separate Google merchant registration required for UCP enablement — none of this evidences an open, versioned, sign-nothing spec.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Access to PayPal's agentic commerce services is form-gated, not self-serve: merchants must complete a form "to contact the AI team at PayPal…”
   • [claimed-docs] “"Agent Ready helps Braintree merchants accept payments from AI shopping assistants across major platforms, including ChatGPT, Google AI Mode…”

   proof ↗⚑ flag

Disputes

1. developerDisputes and chargebacks on agent transactions can be handled via API — evidence submission, status, outcomes

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce—none0/10

   The evidence pack documents refunds via the standard Refunds API/Dashboard for machine payments, and general agent-approval workflows for actions like refunds, but contains no mention of a disputes/chargebacks API, evidence submission, dispute status tracking, or outcome reporting for agent-initiated transactions specifically.

   ⚑ flag

   PayPal Agentic Commerce~partialclaimed6/10

   The agent toolkit exposes dispute tools — list_disputes, get_dispute, and accept_dispute_claim — plus the general REST API surface explicitly lists 'disputes' alongside orders/payments, giving developers programmatic access to dispute status and resolution outcomes. However, no evidence-submission endpoint (e.g., uploading documents/evidence for a dispute) is documented anywhere in the pack, so the 'evidence submission' part of the story is unsupported. missing for 10: documented evidence-submission API/tool for disputes, and any hands-on/independent confirmation that these dispute tools work end-to-end in production.

   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “Integrate with PayPal APIs to access orders, invoices, subscriptions, shipment tracking, transaction details, and dispute management through…”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”

   proof ↗⚑ flag

Refunds

1. developerRefunds on agent-placed orders can be initiated and tracked programmatically

   weight 2 · round to PayPal Agentic Commerce
   Stripe Agentic Commerce~partialclaimed5/10

   Evidence shows refunds are available through Stripe's standard Refunds API/Dashboard (inherited by agent-driven orders) and that the Stripe MCP server explicitly supports refunds as a stripe_api_write action, gated by human approval before an agent can execute or retry it. However, there's no agent-specific refund tracking mechanism (e.g., dedicated refund status webhooks tied to agent_details, or refund reporting in the agent-tagged Transactions view) beyond the standard Stripe rails. Missing for 10: dedicated refund tracking/webhooks scoped to agent orders, explicit API examples for programmatic refund initiation on SPT-based orders, and independent/hands-on confirmation of the refund flow working end-to-end.

   • [claimed-docs] “"Machine payments let agents pay for APIs and services programmatically. Your server returns a payment challenge, the agent presents a valid…”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialprobed6/10

   The Agent Toolkit exposes create_refund/get_refund tools and dispute tools (list_disputes, get_dispute, accept_dispute_claim) so refunds on agent-placed orders can be initiated and tracked programmatically via MCP/LangChain/etc., backed by the underlying PayPal REST orders/disputes API and webhooks for status updates. However, the docs don't specifically address refund flows tied to the new agentic-commerce cart/checkout (ACP/UCP) orders, and there's no independent hands-on confirmation of refund success for agent-originated transactions. missing for 10: explicit documentation linking refund/dispute tools to agent-placed (ACP/UCP/Store Sync) orders specifically, and independent/hands-on verification of the refund tracking flow.

   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the PayPal Agent Toolkit resolves on the public npm registry (npm view @paypal/agent-toolkit → 1.11.0, …”
   • [claimed-docs] “The developer portal's REST API getting-started page anchors the agent surface: "REST APIs, including orders, payments, subscriptions, invoi…”

   proof ↗⚑ flag

Agent identity

1. merchantAgents identify themselves verifiably — registered profiles, signatures, or trust tiers — so I can tell trusted agent traffic from anonymous bots

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed6/10

   Stripe's ACS onboarding shows a real per-agent registration/approval flow ('review its terms' and 'send the agent an approval request that the agent must accept') and an OCA connection that both parties must confirm before commerce flows, plus order-level attribution ('tagged with the originating agent', filterable by agent name) and scoped, revocable SPT grants — all of which let a merchant distinguish known, approved agents from anonymous traffic. However there is no documented cryptographic signature scheme, public agent identity registry, or explicit trust-tier/reputation system merchants can query independently of the approval workflow. Missing for 10: cryptographic agent signatures/verifiable credentials, a public or queryable trust-tier/reputation system, and independent (non-vendor) evidence that merchants can reliably detect spoofed or unregistered agent traffic.

   • [claimed-docs] “Merchant onboarding is Dashboard self-serve — "Use Agentic Commerce Suite (ACS) to start selling through agents with a single integration", …”
   • [claimed-docs] “Agent orders flow through standard Stripe rails with per-agent attribution: "Stripe sends `checkout.session.completed` after the agent compl…”
   • [claimed-docs] “For agent builders, the embedded mode manages "the full checkout lifecycle through Stripe, including product feed ingestion, cart management…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “RAKs are the recommended type of key for API keys you give to AI agents.”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   PayPal's docs describe payment-token authentication (Braintree single-use tokens bound to merchant ID), PayPal-issued JWTs for verifying that requests come from PayPal's own Shopping Cart service, and webhook signature verification — but none of this constitutes a verifiable agent identity system (no registered agent profiles, agent-level signatures, or trust tiers that would let a merchant distinguish which specific AI agent or bot is originating traffic). The authentication documented is merchant-to-PayPal or PayPal-to-merchant channel security, not agent-to-merchant identity/trust signaling.

   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”
   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”
   • [probe] “PROBE runtime (recorded 2026-09-14): PayPal's remote MCP server is live — a keyless JSON-RPC initialize POST to https://mcp.paypal.com/mcp a…”

   proof ↗⚑ flag

Human oversight

1. ai-native userSensitive agent spending requires my documented approval — a confirmation step, signed mandate, or per-purchase consent — before money moves

   weight 3 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce✓fullprobed8/10

   Stripe documents multiple concrete consent gates before money moves: MCP requires explicit human 'Approve' confirmation with an approval token before write actions like refunds/outbound payments (expiring in 24h), Link CLI requires the customer to 'approve each request on the Link website or in the mobile app' before releasing one-time-use credentials, and Shared Payment Tokens carry issuer-set usage_limits (currency, max_amount, expiration) plus revocation and a documented state machine with webhooks. Runtime probes corroborate the MCP OAuth-gated approval flow and the live Link CLI package. missing for 10: independent/hands-on evidence of the approval UX actually blocking a real transaction end-to-end (only docs/runtime probes, no third-party test), and no evidence of a 'signed mandate' artifact akin to AP2's cryptographic mandates.

   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “"Link CLI gives your agent access to a wallet that your customer controls. Your agent can retrieve one-time-use payment credentials to compl…”
   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [probe] “PROBE runtime (recorded 2026-09-14): Stripe's hosted MCP server is live and gated exactly as documented — a keyless JSON-RPC initialize POST…”
   • [probe] “PROBE runtime (recorded 2026-09-14): the Link CLI — the agent-wallet client documented for retrieving one-time-use payment credentials and p…”
   • [claimed-docs] “Customers approve each request on the Link website or in the mobile app.”

   proof ↗⚑ flag

   PayPal Agentic Commerce~partialclaimed5/10

   PayPal's ACP/UCP flows use single-use, time- and amount-bound delegated tokens tied to a specific merchant and transaction, and the toolkit docs explicitly recommend 'human oversight' for judgment-sensitive actions — both suggest transaction-level guardrails rather than fully autonomous spending. However, the evidence never documents an explicit user-facing confirmation screen, signed mandate artifact, or per-purchase consent UI shown to the end user before the token is used to move money; the checkout flow (requestCheckout/Instant Checkout) is described only at a technical/token level, not from the consent-UX perspective. Missing for 10: an explicit documented consent/confirmation step or mandate artifact shown to the payer, and any independent/hands-on confirmation that this consent step actually blocks unauthorized charges.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “"PayPal's agent toolkit supports the integration of PayPal APIs into AI agent workflows using" Amazon Bedrock, CrewAI, LangChain, MCP, OpenA…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”

   proof ↗⚑ flag

Liability

1. merchantThe fraud, liability, and chargeback rules for agent-initiated transactions are documented so I know who bears the risk

   weight 2 · round drawn
   Stripe Agentic Commerce—none0/10

   The evidence pack covers SPT scoping/revocation, checkout session lifecycle, human-approval gates for refunds, and network Agent Pay/Intelligent Commerce token programs, but nowhere states explicit fraud-liability or chargeback-allocation rules (e.g., who bears loss on a disputed agent-initiated charge, liability-shift terms, or chargeback responsibility between merchant/agent/Stripe). This is a fair question for a payments platform (trust-safety axis clearly applies), but no documentation in the pack addresses it.

   ⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence shows general dispute-management and refund tools (list_disputes, accept_dispute_claim, create_refund) and standard PayPal fee/webhook docs, but nothing documents fraud, liability, or chargeback allocation specifically for agent-initiated transactions (e.g., who bears risk when a delegated token or AI agent purchase is disputed). The delegated-token docs mention tokens are 'bound to your merchant ID' but stop short of stating liability shift or chargeback rules for agent-driven checkouts.

   • [claimed-docs] “Post-purchase operations are first-class agent tools: refunds via create_refund / get_refund (a "refund for a captured payment", where "If n…”
   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The published US commercial rates that agent-initiated orders inherit: "PayPal Checkout" and "PayPal Guest Checkout" at "3.49% +" fixed fee,…”
   • [claimed-docs] “"Webhooks are HTTPS posts from PayPal to an endpoint on your server whenever the corresponding event type occurs" — "Up to 10 webhook URLs m…”

   proof ↗⚑ flag

Spend controls

1. ai-native userSet per-agent spend caps and usage limits and revoke an agent's payment authority instantly

   weight 2 · round to Stripe Agentic Commerce
   Stripe Agentic Commerce~partialclaimed7/10

   Shared Payment Tokens explicitly support per-token usage_limits (currency, max_amount, expiration) and can be revoked at any time so sellers can't use them ('Sellers can't create a payment with a revoked SPT'), and Restricted API Keys (RAKs) let a developer scope what an agent's key can do, while MCP write actions require human approval tokens that expire in 24h. However, these are per-token/per-key controls rather than a single unified 'per-agent' spend-cap dashboard, and there's no evidence of one-click instant revocation of an agent's entire payment authority across all rails (card, machine payments, x402, Link wallet) at once. missing for 10: a unified per-agent spend-cap/limit management UI or API (vs. per-SPT/per-key limits), evidence of instantly revoking ALL of an agent's payment methods/authority in one action, and independent/hands-on confirmation that revocation works as documented in production.

   • [claimed-docs] “"Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases." The issuing agent mu…”
   • [claimed-docs] “RAKs are the recommended type of key for API keys you give to AI agents.”
   • [claimed-docs] “Dashboard-side agent actions carry a documented human approval gate: "To prevent agents from making mistakes, Stripe requires human confirma…”
   • [claimed-docs] “Shared payment tokens (SPTs) grant sellers scoped access to customers’ payment methods for agent-initiated purchases.”

   proof ↗⚑ flag

   PayPal Agentic Commerce—none0/10

   Evidence shows per-transaction delegated tokens with configurable amount/time restrictions bound to a merchant (paypal-ac-supp-acp-chatgpt), but nothing about a user or merchant setting persistent per-agent spend caps/usage limits across multiple transactions, or an instant revoke-authority mechanism for a given agent's payment access. No dashboard, API, or docs reference for agent-level authorization management or revocation is present.

   • [claimed-docs] “The ACP path is an in-ChatGPT checkout build: "Implement an MCP server with the `complete_checkout` tool to receive tokens", "Call `requestC…”
   • [claimed-docs] “The UCP path targets Google surfaces: "This enables AI-powered checkout experiences through Google AI Mode (Gemini), where Google Pay return…”
   • [claimed-docs] “Cart execution is a reverse API the merchant hosts: POST /merchant-cart "Creates a new cart with the provided PayPalCart object. This is cal…”

   proof ↗⚑ flag