Penpot vs Claude Design

Agent ops

1. ai-native userAn agent can create or modify designs programmatically — through a write-capable API, plugin surface, or MCP tools — not just read them

   weight 2 · round to Penpot
   Penpot~partialprobed6/10

   Penpot has a documented MCP server explicitly for connecting AI agents/LLMs to design files, plus a plugin API and integrations API with access to objects/properties, suggesting agents can both read and modify designs programmatically. However, MCP docs emphasize agents 'reading and interacting with' design files for high-fidelity artifacts, and no independent/hands-on evidence confirms write operations via MCP or API actually work reliably, and openapi/spec probes 404'd. Missing for 10: concrete documented write/create examples via MCP or API, independent verification of agent-driven design edits, and a working OpenAPI spec.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Connect Penpot MCP server to your trusted AI agent. Customize your models and design infrastructure however works best for your team.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design~partialprobed4/10

   There's no documented first-party write API or MCP server for Claude Design itself (OpenAPI probe returned 404s), but Claude Code slash-commands (/design, /design-sync) and handoff bundles let an agent programmatically create/modify designs, and one community report describes a working MCP flow pushing Claude Design output into PenPot with reusable components. Missing for 10: official documented MCP server/API surface for Claude Design, broader independent corroboration of agent-driven write access beyond a single anecdote.

   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://support.claude.com/openapi.json, https://support.claude.com/swagger.json, https://support.cl…”

   proof ↗⚑ flag

2. ai-native userAn agent can render and export designs — screenshots, images, or production formats — through a documented programmatic surface

   weight 2 · round drawn
   Penpot~partialprobed5/10

   Penpot documents an MCP server and plugin/integrations API that give agents structured access to design objects and mentions 'external rendering' as a use case (penpot-docs-3, penpot-docs-32), and design data can be exported to open formats like SVG/CSS/HTML (penpot-docs-21). However, there is no concrete documented endpoint or workflow for producing screenshots/images/production export files programmatically, no OpenAPI spec was found (404s in penpot-probe-2), and one community report describes limited export options (single-layer only) in hands-on use (penpot-comm-5). missing for 10: an explicit documented render/export API or endpoint (PNG/PDF/JPG) usable by an agent, a working OpenAPI/schema reference, and independent confirmation that MCP/plugin API can actually produce exported artifacts.

   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [community] “Maybe I'm using it wrong... when I tried Penpot about a year ago, I had no way to export the diagram as one file like PDF or JPG. I could on…”

   proof ↗⚑ flag

   Claude Design~partialprobed5/10

   Docs confirm export to PPTX/PDF/HTML and a Claude Code handoff bundle triggered via /design-sync, and community reports an MCP-based flow (into PenPot) suggesting some programmatic surface exists, but there is no documented API/SDK for rendering or exporting designs and the openapi probe returned 404s, indicating no formal machine-callable spec. missing for 10: documented API/SDK for export, evidence of screenshot/image export via a callable interface, corroborated OpenAPI or MCP server documentation.

   • [claimed-docs] “Export to PPTX, PDF, or HTML, share an org-scoped link, hand off to Claude Code, or send to the apps you already use daily.”
   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://support.claude.com/openapi.json, https://support.claude.com/swagger.json, https://support.cl…”

   proof ↗⚑ flag

3. ai-native userMy coding agent can pull real design context — frames, variables, screenshots, component mappings — through an MCP server or API to implement the design

   weight 3 · round to Penpot
   Penpot~partialprobed6/10

   Penpot documents an official MCP server that lets any AI agent connect to design files, described as exposing design as code so agents can 'read and interact with your design files' and access objects/properties for external rendering, plus separate docs on components, design tokens, and design-system alignment (penpot-docs-1,18,27,31,32,3,33,11). However, no evidence explicitly confirms screenshot export or an explicit frame/variable/component mapping schema via the MCP itself, and there is no independent/hands-on report validating the MCP server's real-world behavior (only first-party docs, plus unrelated community comments about general app performance). Missing for 10: independent verification of MCP server functioning as described, explicit mention of screenshot/visual context retrieval, and concrete component-mapping API examples.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files. This can help you maintain consistency across a gr…”
   • [claimed-docs] “Design Tokens - Make design decisions that align designers and developers' workflows.”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design~partialcommunity5/10

   Claude Design ships explicit hand-off paths for coding agents — /design-sync to pull the design system into Claude Code, packaged handoff bundles, and import from GitHub/codebase/component libraries — which functionally lets an agent implement designs with real context. However, the mechanism is documented as slash-command/bundle integration rather than an explicit MCP server or general API exposing frames, variables, screenshots, and component mappings; the one community mention of MCP (comm-7) describes Claude Design consuming an external tool's MCP (PenPot), not exposing its own for coding agents to pull context. missing for 10: first-party documentation of an MCP server or API surface for frames/variables/screenshots/component mappings, and independent verification that agents (not just Claude Code) can pull that structured context.

   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”
   • [claimed-docs] “Import from GitHub, design files, or your local codebase so Claude can build with your real components.”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [community] “I have been using Claude Design + Claude Code, and results have been excellent. The handoff skill in Claude Design is pretty solid.”

   proof ↗⚑ flag

Agent access

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

   weight 2 · round to Penpot
   Penpot✓fullprobed8/10

   Penpot publishes an actual llms.txt at penpot.app/llms.txt with structured content (design tokens, plugin marketplace stats, Figma comparison) as cited in penpot-docs-16 and penpot-docs-24, plus extensive agent-oriented docs for its MCP server (penpot-docs-1, 18, 19, 27, 31, 32, 39) explicitly aimed at AI agents. missing for 10: an independent probe directly confirming penpot.app/llms.txt resolves (the probe only tested the unrelated help.penpot.app subdomain and got 404), and no third-party corroboration that agents successfully consume this file.

   • [claimed-docs] “Penpot vs Figma: Open-source alternative with self-hosting, unlimited files, native design tokens.”
   • [claimed-docs] “Plugins (113+): Community plugin marketplace with accessibility, code, AI, and design system tools.”
   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”
   • [probe] “PROBE llms.txt: HTTP 404 at https://help.penpot.app/llms.txt”

   proof ↗⚑ flag

   Claude Design✓fullprobed7/10

   Probes confirm llms.txt is live (HTTP 200) at the Claude support domain and that Claude Design's own docs page is available in agent-friendly markdown form (docs-md probe), meaning an agent could be pointed at these resources to learn the product. Missing for 10: explicit vendor documentation describing llms.txt as an intended agent-consumption channel, and community/hands-on confirmation that an agent actually used llms.txt to complete a task with Claude Design.

   • [probe] “PROBE llms.txt: HTTP 200 at https://support.claude.com/llms.txt # Claude Help Center > Search for answers or browse by topic ## English #…”
   • [probe] “PROBE docs-md: HTTP 200 at https://support.claude.com/en/articles/14604416-get-started-with-claude-design.md # Get started with Claude Desig…”

   proof ↗⚑ flag

2. ai-native userPlug MCP servers into this product so it can use their tools

   weight 3 · round drawn
   Penpot~partialclaimed3/10

   Penpot's docs mention connecting to another product's MCP server ('Migrate your designs into Penpot by connecting Figma's and Penpot's MCPs') implying Penpot can act as an MCP client to import tool-provided data, but the vast majority of MCP evidence describes Penpot instead exposing its OWN MCP server for agents to connect to (the reverse relationship), not Penpot consuming arbitrary external MCP servers' tools. Missing for 10: documentation of a general-purpose MCP client/connector UI in Penpot, a list of supported external MCP servers beyond Figma, and any hands-on/community confirmation that this client-side integration works.

   • [claimed-docs] “Migrate your designs into Penpot by connecting Figma’s and Penpot’s MCPs.”
   • [claimed-docs] “Migrate your designs into Penpot by connecting Figma's and Penpot's MCPs.”
   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic”

   proof ↗⚑ flag

   Claude Design~partialcommunity3/10

   Only a single ambiguous community mention notes 'Claude Design into PenPot via its MCP,' suggesting some MCP-based integration exists, and docs reference pulling data from tools like Salesforce/QuickBooks/HubSpot without specifying MCP as the mechanism. There is no official documentation confirming users can plug arbitrary MCP servers into Claude Design to expose their tools. Missing for 10: first-party docs on MCP server support, clarity on directionality (client vs server), and broader independent confirmation beyond one ambiguous forum comment.

   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [claimed-docs] “Turn this proposal into a one-pager a prospect will want to open. Pull their details from Salesforce and export it as a PDF.”
   • [claimed-docs] “Dig into your numbers in Intuit QuickBooks, plan a promotion to get sales up, generate branded assets in Canva, and get your next campaign s…”

   proof ↗⚑ flag

3. ai-native userUse an official CLI

   weight 2 · round drawn
   Penpot—none0/10

   Evidence covers Penpot's MCP server, plugin API, webhooks, and REST access tokens, but there is no mention of an official CLI tool for interacting with Penpot as an AI-native workflow interface. The MCP server is a distinct integration mechanism, not a CLI.

   ⚑ flag

   Claude Design—none0/10

   Evidence shows Claude Design integrates with Claude Code via slash commands (/design-sync, /design) and handoff bundles, but this is Claude Code's CLI, not an official CLI shipped by Claude Design itself; no standalone CLI for Claude Design is documented.

   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”

   proof ↗⚑ flag

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

   weight 3 · round to Penpot
   Penpot✓fullprobed8/10

   Penpot documents a public API with access tokens for authentication, outbound webhooks, an open plugin API, and a dedicated MCP server explicitly designed to let any AI agent/LLM read and manipulate design files programmatically. This directly matches the ai-native 'documented public API' story, with self-hosting options adding control. Missing for 10: no discoverable OpenAPI/swagger spec (probes returned 404s) and no independent/hands-on verification of the API's completeness beyond vendor docs.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “Use access tokens as an alternative login authentication. They can allow an app to access Penpot's API.”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design—none0/10

   Evidence pack shows only end-user product docs (import/export, design generation, handoff to Claude Code) and no documented public API; a direct probe for OpenAPI/swagger specs at claude.com/support returned 404s, and no REST/SDK API for Claude Design is mentioned anywhere.

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

   proof ↗⚑ flag

5. ai-native userBuild against official SDKs

   weight 2 · round to Penpot
   Penpot✓fullprobed7/10

   Penpot ships an official, self-hostable MCP server plus an open plugin API, REST API with access-token auth, and webhooks, explicitly marketed for AI agents to build against (penpot-docs-1,3,5,6,19,27,31,32,39). This gives AI-native users concrete official SDK/API surfaces to integrate with, though there's no independent hands-on validation of the MCP server and the OpenAPI spec could not be located (probe 404s), missing for 10: independent developer confirmation of MCP/plugin SDK reliability, discoverable machine-readable OpenAPI/llms.txt spec.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “Our plugin system brings the flexibility that devs need to create and publish their own custom extensions.”
   • [claimed-docs] “Use access tokens as an alternative login authentication. They can allow an app to access Penpot's API.”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Connect Penpot MCP server to your trusted AI agent. Customize your models and design infrastructure however works best for your team.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design—none0/10

   Claude Design's evidence pack shows only end-user product docs (mockups, prototypes, handoff, exports) and community discussion of MCP integration with third-party tools like PenPot, but no mention of an official SDK for building against Claude Design itself. A direct probe for API/OpenAPI specs returned 404s, indicating no public developer SDK is documented.

   • [probe] “PROBE openapi: all candidate paths 404 (https://support.claude.com/openapi.json, https://support.claude.com/swagger.json, https://support.cl…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”

   proof ↗⚑ flag

Agentic features

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

   weight 2 · round drawn
   Penpot~partialclaimed5/10

   Penpot exposes design data via its MCP server, plugin API, and webhooks so external AI agents can read structured design data and generate outputs (e.g., 'AI agents can easily read and interact with your design files creating high-fidelity artifacts') and the plugin marketplace lists AI tools, but there is no evidence of a built-in, first-party AI feature that proactively surfaces insights or suggestions from a user's own data inside the product itself. Missing for 10: a native in-app AI insight/suggestion feature (vs. only enabling third-party agents via MCP/plugins), and any hands-on/community confirmation that such insights actually work in practice.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Penpot enables plenty of AI workflows, from generating editable design structures to connecting product logic, tokens, and implementation.”
   • [claimed-docs] “Plugins (113+): Community plugin marketplace with accessibility, code, AI, and design system tools.”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”

   proof ↗⚑ flag

   Claude Design~partialclaimed5/10

   Claude Design can pull in external data (Salesforce, QuickBooks, GitHub) and use it to generate on-brand designs, proposals, or campaign plans, which is a limited form of data-driven suggestion, but this is framed around design generation rather than analytical insight delivery, and no evidence shows dedicated 'insights/suggestions' features from uploaded datasets. missing for 10: dedicated data-analysis/insight generation UI, independent corroboration that it surfaces analytical insights (not just design outputs) from imported data.

   • [claimed-docs] “Turn this proposal into a one-pager a prospect will want to open. Pull their details from Salesforce and export it as a PDF.”
   • [claimed-docs] “Dig into your numbers in Intuit QuickBooks, plan a promotion to get sales up, generate branded assets in Canva, and get your next campaign s…”
   • [claimed-docs] “Import from GitHub, design files, or your local codebase so Claude can build with your real components.”
   • [claimed-docs] “Import from your codebase, a web capture, or DOCX, PPTX, and XLSX files.”

   proof ↗⚑ flag

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

   weight 2 · round to Penpot
   Penpot~partialclaimed4/10

   Penpot provides building blocks for automation—an MCP server for AI agents, outbound webhooks, and a plugin API—that could underlie background automations, but all documented use cases (prompting an agent, plugin actions) are interactive/triggered rather than autonomous, unattended background processes. Missing for 10: explicit support for scheduled/triggered autonomous agent runs, evidence of agents operating without a live user prompt, and any independent confirmation that MCP/webhook flows work unattended in production.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Outbound webhooks are event calls from Penpot to other applications. They notify that some event occurred, e.g. a file has been created or u…”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “Penpot enables plenty of AI workflows, from generating editable design structures to connecting product logic, tokens, and implementation.”

   proof ↗⚑ flag

   Claude Design—none0/10

   Claude Design is a conversational/interactive design tool for generating mockups, prototypes, and handoffs; nothing in the evidence describes scheduled tasks, triggers, or autonomous background runs — all workflows are synchronous and user-initiated via chat or explicit commands.

   ⚑ flag

3. ai-native userDelegate tasks to a built-in AI assistant inside the product

   weight 3 · round to Claude Design
   Penpot—none0/10

   Penpot's AI evidence is entirely about an MCP server that lets external AI agents/LLMs connect to Penpot's data (penpot-docs-1, penpot-docs-18, penpot-docs-27, penpot-docs-31, penpot-docs-39), not about a built-in AI assistant embedded in the product itself that users can delegate tasks to. No evidence describes an in-app AI chat/copilot feature within the Penpot UI.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “Connect Penpot MCP server to your trusted AI agent. Customize your models and design infrastructure however works best for your team.”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design✓fullcommunity8/10

   Claude Design is built around delegating design tasks directly to Claude within the product—describing a visual, refining through conversation/sliders, generating variations, and packaging handoffs are all done by the embedded Claude assistant, and this is corroborated by hands-on community reports of building shippable pages and using the handoff skill successfully. missing for 10: independent quality benchmarking of how much genuine agentic reasoning vs templated generation occurs, and some community reports note underwhelming or 'boring' outputs suggesting delegation quality varies.

   • [claimed-docs] “Describe a visual and Claude builds a first version, on brand.”
   • [claimed-docs] “Refine, share, or hand off to the tools you use. You're the designer, start to finish.”
   • [claimed-docs] “you refine through conversation, inline comments, direct edits, or custom sliders (made by Claude) until it's right”
   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [community] “Used Claude Design to build the landing page for my side project over the weekend. Got something shippable in a few hours. It's great for 80…”
   • [community] “I have been using Claude Design + Claude Code, and results have been excellent. The handoff skill in Claude Design is pretty solid.”
   • [community] “I tested it yesterday. Kinda impressive, but also design output is pretty boring.”

   proof ↗⚑ flag

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

   weight 2 · round to Claude Design
   Penpot✓fullprobed7/10

   Penpot ships an official MCP server explicitly designed so any AI agent/LLM can 'prompt away' to read and manipulate design files expressed as code, directly enabling natural-language operation, backed by plugin API and access tokens for automation. Missing for 10: independent/hands-on evidence of the MCP server actually working end-to-end in practice (community comments only cover general UI/performance, not the AI/MCP workflow specifically).

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Connect Penpot MCP server to your trusted AI agent. Customize your models and design infrastructure however works best for your team.”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design✓fullcommunity8/10

   Claude Design's entire interaction model is natural-language: users 'describe a visual and Claude builds a first version,' refine 'through conversation, inline comments,' and use conversational handoffs like /design and /design-sync — with real-world use (landing pages, PenPot flows) confirming this works in practice, though output quality and robustness vary across users. Missing for 10: consistent quality/benchmarks across users and no independent rigor testing beyond anecdotal HN reports.

   • [claimed-docs] “Describe a visual and Claude builds a first version, on brand.”
   • [claimed-docs] “you refine through conversation, inline comments, direct edits, or custom sliders (made by Claude) until it's right”
   • [claimed-docs] “Now you can use Claude Design from inside any conversation with Claude.”
   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [community] “Used Claude Design to build the landing page for my side project over the weekend. Got something shippable in a few hours. It's great for 80…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [community] “I tested it yesterday. Kinda impressive, but also design output is pretty boring.”

   proof ↗⚑ flag

Api quality

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

   weight 2 · round drawn
   Penpot—none0/10

   While Penpot documents a general integrations API, access tokens, webhooks, and an MCP server, there is no evidence of a downloadable OpenAPI/Swagger spec; direct probes for openapi.json/swagger.json/llms.txt all returned 404s, indicating no machine-readable API spec is published.

   • [probe] “PROBE llms.txt: HTTP 404 at https://help.penpot.app/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “Use access tokens as an alternative login authentication. They can allow an app to access Penpot's API.”

   proof ↗⚑ flag

   Claude Design—none0/10

   Claude Design is a design/prototyping product with no evidence of a public API, and the direct probe for OpenAPI/swagger endpoints returned 404 on all candidate paths, with no other documentation mentioning a machine-readable API spec.

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

   proof ↗⚑ flag

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

   weight 1 · round to Claude Design
   Penpot~partialclaimed4/10

   Penpot has no dedicated 'sandbox' or 'test environment' feature, but its self-hosting capability (docker-compose/Kubernetes deployment, MCP server deployable in your own environment) means an AI-native user could spin up an isolated instance separate from production to test agent workflows. This is an indirect workaround rather than a purpose-built sandbox/staging mode. Missing for 10: an explicit sandbox/staging environment feature, documented separation of test vs production data within the same account, and any first-party guidance on safely testing AI agents against non-production data.

   • [claimed-docs] “Run it in your private cloud, on-premise, or any environment you control.”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”
   • [claimed-docs] “you can set up an instance of Penpot in just 3 minutes!”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”
   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”

   proof ↗⚑ flag

   Claude Design~partialclaimed5/10

   Claude Design's core workflow is prototyping (mockups, interactive prototypes) that are explicitly separate from production code — 'no PRs, no code review' before user-testing — which implies a sandbox-like separation from shipped/production systems. However, there is no explicit mention of a dedicated sandbox environment, test data isolation, or protection of 'production data' in a technical/backend sense; the framing is about design/code review gating rather than data safety. missing for 10: explicit sandbox/staging environment terminology, guarantees about not touching production data or backend systems, and independent confirmation of this isolation in practice.

   • [claimed-docs] “Turn a static mockup into a shareable, interactive prototype you can test with real people. No PRs, no code review.”
   • [claimed-docs] “Designers can turn static mockups into easily-shareable interactive prototypes to gather feedback and user-test, without code review or PRs.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”

   proof ↗⚑ flag

Ai

1. designerBuilt-in AI handles asset chores — image generation, background removal, content-aware fills, copy suggestions — inside the editor

   weight 1 · round to Claude Design
   Penpot—none0/10

   Penpot's AI evidence is entirely about an MCP server for connecting external AI agents/LLMs to structured design data, plus a generic plugin marketplace mention of 'AI tools' — there is no evidence of built-in editor features like image generation, background removal, content-aware fill, or copy suggestions happening inside Penpot itself. The axis is fair for a design tool (competitors ship such features), but nothing in the pack shows Penpot delivering it natively.

   • [claimed-docs] “Connect Penpot with any AI agent, any LLM. Penpot MCP server is agnostic and enables high-quality multi-directional AI workflows through our…”
   • [claimed-docs] “Penpot connects any AI agent with structured design. Automate tedious tasks, streamline your processes between apps, and build truly powerfu…”
   • [claimed-docs] “Plugins (113+): Community plugin marketplace with accessibility, code, AI, and design system tools.”
   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Penpot enables plenty of AI workflows, from generating editable design structures to connecting product logic, tokens, and implementation.”

   proof ↗⚑ flag

   Claude Design~partialclaimed3/10

   Claude Design can generate full visual designs from a text prompt and offers layout/color adjustment sliders (claude-design-docs-13, claude-design-docs-9), which loosely maps to 'image generation' and some asset tuning, but there is no evidence of dedicated background-removal, content-aware fill, or copy-suggestion tools built into the editor. missing for 10: background removal tool, content-aware fill/inpainting, in-editor copy/headline suggestion feature.

   • [claimed-docs] “Describe a visual and Claude builds a first version, on brand.”
   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [claimed-docs] “you refine through conversation, inline comments, direct edits, or custom sliders (made by Claude) until it's right”

   proof ↗⚑ flag

2. designerGenerate or edit designs from a natural-language prompt — first drafts, layouts, or full mockups the tool actually produces

   weight 2 · round to Claude Design
   Penpot~partialprobed5/10

   Penpot's docs describe an official MCP server that lets external AI agents/LLMs read and interact with Penpot files to 'generate editable design structures' and 'create high-fidelity artifacts' from prompts (penpot-docs-31, penpot-docs-32, penpot-docs-40), which supports the story's intent of prompt-driven design generation. However, this is achieved only via connecting a third-party AI agent through MCP rather than a native in-app 'type a prompt, get a mockup' feature, and there is no independent/hands-on evidence confirming actual generated layouts or mockups working in practice. Missing for 10: first-party demo/screenshots of actual generated designs, independent hands-on reports validating the MCP-to-mockup pipeline, and a native in-app generation UI not dependent on external agent setup.

   • [claimed-docs] “Choose your agent, connect to Penpot MCP and prompt away!”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Penpot enables plenty of AI workflows, from generating editable design structures to connecting product logic, tokens, and implementation.”
   • [claimed-docs] “Connect it to any AI agent or LLM you trust. Customize your models and design infrastructure however works best for your team.”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design~partialcommunity6/10

   First-party docs and community reports confirm Claude Design generates first drafts, multiple directions, layouts, and mockups from natural-language prompts with editing/refinement controls (docs-13,20,9,21; comm-6,7,16), and users describe shippable landing pages and solid handoff. However hands-on community feedback also reports underwhelming, 'boring', or 'useless' results for some prompts and generic upload errors (comm-3,9,10,8), showing real quality variance rather than a uniformly reliable draft-to-mockup pipeline. Missing for 10: consistent independent benchmarking of output quality across prompt types, and resolution of reported failure cases (large file uploads, simple prompts underwhelming).

   • [claimed-docs] “Describe a visual and Claude builds a first version, on brand.”
   • [claimed-docs] “Designers can quickly create a wide range of directions to explor”
   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [claimed-docs] “you refine through conversation, inline comments, direct edits, or custom sliders (made by Claude) until it's right”
   • [community] “Used Claude Design to build the landing page for my side project over the weekend. Got something shippable in a few hours. It's great for 80…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”
   • [community] “I have been using Claude Design + Claude Code, and results have been excellent. The handoff skill in Claude Design is pretty solid.”
   • [community] “After having high expectations from watching the intro, the actual result of a simple prompt 'Bear on a bicycle' is very underwhelming.”
   • [community] “I tested it yesterday. Kinda impressive, but also design output is pretty boring.”
   • [community] “I've tried it. It's useless.”

   proof ↗⚑ flag

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

   weight 2 · round to Claude Design
   Penpot!disputedcontradicted5/10

   Penpot's API/plugin/MCP layer gives extensive access to objects and properties and lets agents 'automate tedious tasks' across files (penpot-docs-2, penpot-docs-3, penpot-docs-32), which implies scriptable bulk edits, but no documentation specifically demonstrates a bulk-operation workflow (e.g., batch renaming/restyling hundreds of layers). Independent hands-on reports directly contradict smooth handling of many items: performance 'starts dropping rapidly once you approach around 1000 layers' versus design systems that run 10-40k layers (penpot-comm-19), and users report crashes/lag with just a few pages of content (penpot-comm-10, penpot-comm-16). Missing for 10: a documented bulk-edit API/plugin example, benchmarks showing many-item operations succeed at scale, and resolution of the reported performance degradation.

   • [claimed-docs] “Penpot connects any AI agent with structured design. Automate tedious tasks, streamline your processes between apps, and build truly powerfu…”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [community] “Currently Penpot's performance starts dropping rapidly once you approach around 1000 layers. Most robust design systems I see run around 10-…”
   • [community] “Unstable, very crash prone with just a few users designing 10 plus pages. And a huge memory hog too. I run it on Dedicated server with 64GB …”
   • [community] “I want to like penpot, but on even my beefiest computers it causes the whole system to slow to a crawl when opening anything complex.”

   proof ↗⚑ flag

   Claude Design~partialclaimed4/10

   Claude Design supports generating multiple design directions at once (claude-design-docs-2) and applying a design system consistently 'to every project automatically' (claude-design-docs-3), which suggests some batch-like automation, but there's no evidence of true bulk operations across many items (e.g., batch editing dozens of files/assets, bulk export/import at scale, or programmatic multi-item processing). missing for 10: evidence of bulk/batch processing across many discrete items (not just generating variants), bulk edit/export APIs, scale limits, and independent confirmation of multi-item automation workflows.

   • [claimed-docs] “Generate a dozen directions in minutes, then narrow down in your primary tool.”
   • [claimed-docs] “Claude can also apply your team's design system to every project automatically, so the output is consistent with the rest of your company's …”

   proof ↗⚑ flag

2. ai-native userVersion, review, and roll back my automations

   weight 1 · round drawn
   Penpot—none0/10

   Penpot's evidence covers MCP/AI agent integration, webhooks, and plugins for automation, but there is no evidence of any mechanism to version, review, or roll back these automations themselves (e.g., automation history, audit trail, or undo for agent-driven actions). Component 'restore main' (docs-23) is a design-object rollback, not an automation/workflow rollback feature.

   ⚑ flag

   Claude Design—none0/10

   The evidence pack describes Claude Design's prototyping, design-system sync, and handoff features, but contains no mention of version history, review workflows, or rollback capability for any automations or design iterations it produces. Community comments even note lock-out of prior work after unsubscribing, with no rollback/versioning remedy mentioned.

   ⚑ flag

Collaboration

1. product managerComment directly on designs, mention teammates, and resolve threads so review happens in the file, not in email

   weight 2 · round to Claude Design
   Penpot~partialclaimed2/10

   The only evidence touching commenting is a passing mention that webhooks fire when 'a comment has been added,' which confirms Penpot has some in-file commenting feature but gives no detail on @mentions or resolving threads. missing for 10: documentation or hands-on evidence of @mention functionality, thread resolution UI, and any community confirmation that review workflows actually replace email in practice.

   • [claimed-docs] “Outbound webhooks are event calls from Penpot to other applications. They notify that some event occurred, e.g. a file has been created or u…”
   • [claimed-docs] “Outbound webhooks are event calls from Penpot to other applications. They notify that some event occurred, e.g. a file has been created or u…”

   proof ↗⚑ flag

   Claude Design~partialclaimed3/10

   Evidence shows collaborative editing (grant edit access, group conversation) and 'inline comments' as a refinement mechanism, but there's no documented @mention-teammates feature or thread-resolution workflow that would replace email-based review. Missing for 10: explicit @mention of teammates, comment thread resolution/status tracking, and any hands-on confirmation that commenting works as a review tool rather than just a chat refinement input.

   • [claimed-docs] “grant edit access so colleagues can modify the design and chat with Claude together in a group conversation”
   • [claimed-docs] “you refine through conversation, inline comments, direct edits, or custom sliders (made by Claude) until it's right”

   proof ↗⚑ flag

2. designerMy whole team can edit the same file simultaneously with live cursors and instant sync

   weight 3 · round to Claude Design
   Penpot!disputedcontradicted4/10

   Vendor marketing implies real-time team collaboration ("Collaborate freely with unlimited teams") but the evidence pack contains no explicit documentation of live cursors or instant sync mechanics, and multiple hands-on reports describe multiplayer editing as crash-prone or unstable when several users work simultaneously on a file. Missing for 10: first-party docs describing live cursor presence/conflict resolution, latency/sync architecture details, and independent confirmation that simultaneous multi-user editing performs reliably at scale.

   • [claimed-docs] “Unlimited Teams - Collaborate freely with unlimited teams.”
   • [claimed-docs] “No file limits. Create as many as you need within your storage capacity. ... Unlimited Teams. Collaborate freely with unlimited teams.”
   • [community] “Unstable, very crash prone with just a few users designing 10 plus pages. And a huge memory hog too. I run it on Dedicated server with 64GB …”
   • [community] “I tried to self host penpot a few months ago but the app would crash after a few minutes and not properly show the canvases. So a no for me.”
   • [community] “I really wanted to like penpot, but when I tried a few months ago, simply navigating between pages (even on the example documents) was causi…”

   proof ↗⚑ flag

   Claude Design~partialclaimed4/10

   Docs mention granting edit access so colleagues can modify a design and chat with Claude in a group conversation, implying some shared editing, but there is no evidence of live cursors, simultaneous multi-user editing, or instant real-time sync mechanics typical of multiplayer design tools. Missing for 10: explicit live-cursor/presence indicators, real-time simultaneous editing demonstration, and independent confirmation of true multiplayer sync.

   • [claimed-docs] “grant edit access so colleagues can modify the design and chat with Claude together in a group conversation”

   proof ↗⚑ flag

Design systems

1. designerManage design tokens or variables — colors, spacing, typography — with modes or themes applied across designs

   weight 2 · round to Penpot
   Penpot~partialclaimed4/10

   Penpot explicitly advertises 'Design Tokens' and being a 'native design tokens' tool (penpot-docs-33, penpot-docs-13, penpot-docs-16), which directly supports token management for design-system alignment, but the evidence is only marketing tagline text with no detail on how modes/themes, multiple token sets, or per-property (color/spacing/typography) switching actually work. Components with linked instances (penpot-docs-11, penpot-docs-22) support consistency but aren't tokens/modes specifically, and no community/hands-on evidence corroborates the tokens feature's real-world robustness. Missing for 10: detailed documentation of token modes/themes switching, per-property (color/spacing/typography) token examples, and independent/hands-on confirmation that the tokens feature works as described.

   • [claimed-docs] “Design Tokens - Make design decisions that align designers and developers' workflows.”
   • [claimed-docs] “Make design decisions that align designers and developers' workflows.”
   • [claimed-docs] “Penpot vs Figma: Open-source alternative with self-hosting, unlimited files, native design tokens.”
   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files. This can help you maintain consistency across a gr…”
   • [claimed-docs] “All component copies used in a file are linked in a way that updates made to the Main component can reflect in their component copies.”

   proof ↗⚑ flag

   Claude Design~partialcommunity3/10

   Claude Design can 'apply your team's design system to every project automatically' and offers sliders to tune spacing/color, and one community report describes it building 'reusable PenPot components and design system tokens' via an MCP export flow — but there's no first-party evidence of native token/variable definitions with mode/theme switching (light/dark, brand variants) inside Claude Design itself. missing for 10: explicit token/variable editor, multi-mode/theme support, first-party documentation of applying themes across multiple designs.

   • [claimed-docs] “Claude can also apply your team's design system to every project automatically, so the output is consistent with the rest of your company's …”
   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”

   proof ↗⚑ flag

2. designerBuild reusable components with variants and per-instance overrides that stay linked to their source

   weight 3 · round to Penpot
   Penpot~partialcommunity5/10

   Docs confirm reusable components with main/copy linkage that propagates updates (docs-11,22,23,35) and mention 'variants' as an extension (docs-41), but per-instance override mechanics are never detailed, and hands-on community feedback explicitly flags that variants/states handling is 'not on Figma's level yet' and components can feel 'difficult to handle' (penpot-comm-1, penpot-comm-9). missing for 10: documented per-instance override workflow, detailed variant creation/management docs, independent confirmation that variant/override linkage works robustly at scale.

   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files. This can help you maintain consistency across a gr…”
   • [claimed-docs] “All component copies used in a file are linked in a way that updates made to the Main component can reflect in their component copies.”
   • [claimed-docs] “If a main component has been deleted and you have access to a copy of it, you can use the copy to restore its main.”
   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files.”
   • [claimed-docs] “Extend components with variants—simpler to use, easier to control.”
   • [community] “Good: It is a good copy of figma's UI... Bad: In design systems particularly at variants and states of components, it is not on figma's leve…”
   • [community] “I've been toying a little with self hosted Penpot and it was really nice... Penpot was easy to use though it felt a little difficult to hand…”

   proof ↗⚑ flag

   Claude Design—none0/10

   Claude Design is described as a prompt-driven design/prototyping tool that can 'apply your team's design system' and import components, but there is no evidence of a structured component system with defined variants and per-instance overrides that stay linked to a master source, as in tools like Figma. The one related community mention (comm-7) is about exporting to PenPot's component system, not a native capability of Claude Design itself.

   • [claimed-docs] “Claude can also apply your team's design system to every project automatically, so the output is consistent with the rest of your company's …”
   • [claimed-docs] “Import from GitHub, design files, or your local codebase so Claude can build with your real components.”
   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”

   proof ↗⚑ flag

3. designerShared libraries publish components and styles across files and projects, with controlled updates when the source changes

   weight 2 · round to Penpot
   Penpot~partialcommunity5/10

   Docs confirm components as reusable linked elements where main-component edits propagate to copies (penpot-docs-22, penpot-docs-23, penpot-docs-35) and design-system features like tokens and variants (penpot-docs-33, penpot-docs-41), but there is no explicit documentation of publishing a library across separate files/projects or how updates propagate cross-project, and community feedback notes design-system/component sophistication lags Figma (penpot-comm-1, penpot-comm-9). missing for 10: explicit shared/library publishing workflow across files and projects, documentation of controlled update/accept-changes flow at library level, independent confirmation of cross-project sync reliability.

   • [claimed-docs] “All component copies used in a file are linked in a way that updates made to the Main component can reflect in their component copies.”
   • [claimed-docs] “If a main component has been deleted and you have access to a copy of it, you can use the copy to restore its main.”
   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files.”
   • [claimed-docs] “Design Tokens - Make design decisions that align designers and developers' workflows.”
   • [claimed-docs] “Extend components with variants—simpler to use, easier to control.”
   • [community] “Good: It is a good copy of figma's UI... Bad: In design systems particularly at variants and states of components, it is not on figma's leve…”
   • [community] “I've been toying a little with self hosted Penpot and it was really nice... Penpot was easy to use though it felt a little difficult to hand…”

   proof ↗⚑ flag

   Claude Design~partialclaimed3/10

   Claude Design can apply a team's design system automatically across projects and sync design-system context via '/design-sync' from Claude Code, suggesting some cross-project consistency, but there is no evidence of a true shared-library mechanism (publishing reusable components/styles with versioned, controlled updates propagating when a source library changes) as in dedicated design tools. Missing for 10: library publishing workflow, update/versioning notifications when source components change, and independent confirmation of cross-project sync behavior.

   • [claimed-docs] “Claude can also apply your team's design system to every project automatically, so the output is consistent with the rest of your company's …”
   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”

   proof ↗⚑ flag

Handoff

1. developerDesigns map to production code — component-to-code mappings or direct design-to-code output I can actually ship

   weight 2 · round to Claude Design
   Penpot~partialclaimed6/10

   Penpot documents code-based design representation and an 'Inspect' feature exporting CSS/HTML/SVG/JSON that matches the design (penpot-docs-21/36), plus an MCP server letting AI agents read structured design files to generate implementation artifacts (penpot-docs-32, penpot-docs-40). However, this is inspection/AI-assisted code generation rather than a documented direct component-to-framework code mapping (e.g., to React/Vue components), and no independent/hands-on evidence confirms developers actually shipping this code in production. missing for 10: first-party or community evidence of production-ready component-to-code bindings (framework-specific), and independent verification that generated/inspected code is usable as-is in shipped products.

   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”
   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON”
   • [claimed-docs] “In Penpot, design is expressed as code, so AI agents can easily read and interact with your design files creating high-fidelity artifacts.”
   • [claimed-docs] “Penpot enables plenty of AI workflows, from generating editable design structures to connecting product logic, tokens, and implementation.”
   • [github] “developers feel at home using Penpot, because design is expressed as code, enabling a direct translation and shipping products faster.”

   proof ↗⚑ flag

   Claude Design~partialcommunity7/10

   Docs describe concrete design-to-code handoff: importing real components from GitHub/codebase, syncing design systems via /design-sync, and packaging a 'handoff bundle' passed to Claude Code with one instruction (claude-design-docs-5,6,7,8). Community corroborates this works well ('handoff skill in Claude Design is pretty solid', 'Got something shippable in a few hours') but also flags that the generated code is often unmaintainable ('write-once, read-never' code) and animation/export sanitization bugs, showing real shippability caveats. Missing for 10: independent verification of true component-to-code mapping fidelity (not just handoff bundling) and resolution of maintainability/code-quality concerns raised by users.

   • [claimed-docs] “Import from GitHub, design files, or your local codebase so Claude can build with your real components.”
   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”
   • [community] “I have been using Claude Design + Claude Code, and results have been excellent. The handoff skill in Claude Design is pretty solid.”
   • [community] “It makes absolutely beautiful designs, stunningly so, but the actual code is not something a human could ever maintain. Write-once, read-nev…”
   • [community] “Used Claude Design to build the landing page for my side project over the weekend. Got something shippable in a few hours. It's great for 80…”

   proof ↗⚑ flag

2. developerInspect any layer and get measurements, styles, assets, and code snippets (CSS/iOS/Android) without an editor seat

   weight 3 · round to Penpot
   Penpot~partialclaimed5/10

   Penpot's docs confirm an inspect feature giving code that matches design (CSS, HTML, SVG, JSON), which covers measurements/styles/code snippets for web, but there is no evidence of iOS/Android-specific code snippet generation, and no explicit documentation confirming that viewers/inspectors can access this without a paid editor seat. missing for 10: iOS/Android code snippet support, explicit seat-free viewer/inspect access confirmation, asset-export detail in inspect mode.

   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”
   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON”
   • [claimed-docs] “Unlimited Teams - Collaborate freely with unlimited teams.”
   • [claimed-docs] “No file limits - Create as many as you need within your storage capacity.”

   proof ↗⚑ flag

   Claude Design—none0/10

   Evidence shows Claude Design can export handoff bundles and code (HTML/CSS/JS) to Claude Code or as PPTX/PDF/HTML, but there is no mention of a layer-level 'inspect' mode giving measurements, styles, assets, and platform-specific (CSS/iOS/Android) code snippets for viewers without an editor seat — the axis is applicable to a design/dev-handoff tool but unevidenced here.

   • [claimed-docs] “When a design is ready to build, Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction…”
   • [claimed-docs] “Export to PPTX, PDF, or HTML, share an org-scoped link, hand off to Claude Code, or send to the apps you already use daily.”
   • [community] “It makes absolutely beautiful designs, stunningly so, but the actual code is not something a human could ever maintain. Write-once, read-nev…”

   proof ↗⚑ flag

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

   weight 2 · round to Penpot
   Penpot~partialprobed5/10

   Penpot documents extensive API/plugin/MCP access (objects and properties, webhooks, access tokens, plugin system) suggesting broad programmatic control, but there's no explicit claim or evidence of full UI/API parity, and probes for an OpenAPI/swagger spec returned 404s, indicating the API surface isn't fully discoverable or documented as complete. missing for 10: explicit statement of full UI-API feature parity, a public OpenAPI/schema reference, and independent (community/hands-on) confirmation that all UI actions are scriptable via API.

   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [claimed-docs] “Our plugin system brings the flexibility that devs need to create and publish their own custom extensions.”
   • [claimed-docs] “Use access tokens as an alternative login authentication. They can allow an app to access Penpot's API.”
   • [claimed-docs] “Plugins (113+): Community plugin marketplace with accessibility, code, AI, and design system tools.”
   • [probe] “PROBE llms.txt: HTTP 404 at https://help.penpot.app/llms.txt”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.penpot.app/openapi.json, https://help.penpot.app/swagger.json, https://help.penpot.app/…”
   • [probe] “official MCP server documented at https://penpot.dev/ai/mcp-server”

   proof ↗⚑ flag

   Claude Design—none0/10

   No evidence of a Claude Design API; the only API-related probe found openapi.json/swagger.json endpoints all returning 404, and no docs mention programmatic access to Design's prototyping, handoff, or editing features. All capabilities (prototyping, handoff, design-system sync) are described only as UI/Claude Code slash-command workflows.

   • [probe] “PROBE openapi: all candidate paths 404 (https://support.claude.com/openapi.json, https://support.claude.com/swagger.json, https://support.cl…”
   • [claimed-docs] “Pull in your design system from Claude Code using /design-sync or work directly in Claude Code with /design.”
   • [claimed-docs] “Now you can use Claude Design from inside any conversation with Claude.”

   proof ↗⚑ flag

2. ai-native userExport all of my data in open formats and leave

   weight 3 · round to Penpot
   Penpot!disputedcontradicted5/10

   Penpot's docs promise open formats and no lock-in — the file format is open and design data can be inspected/exported as CSS, HTML, SVG and JSON, and self-hosting gives full data control (penpot-docs-21, penpot-docs-28, penpot-docs-38). However, a hands-on community report directly contradicts full-data export, stating the user 'had no way to export the diagram as one file like PDF or JPG. I could only export a layer' (penpot-comm-5), suggesting the practical export experience falls short of the 'export all my data and leave' promise. Missing for 10: independent confirmation of a full-project/all-files bulk export, verification that the open file format round-trips completely outside Penpot, and resolution of the community-reported export limitation.

   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”
   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”
   • [community] “Maybe I'm using it wrong... when I tried Penpot about a year ago, I had no way to export the diagram as one file like PDF or JPG. I could on…”

   proof ↗⚑ flag

   Claude Design!disputedcontradicted4/10

   Docs confirm export to open-ish formats (PPTX, PDF, HTML) and that designs are HTML/CSS/JS editable by any tool, suggesting an open-format export path. However, community reports document a concrete failure: users lost access to their entire design history/projects after unsubscribing, effectively blocking 'leaving' with their data, which Anthropic staff acknowledged as a bug needing a fix rather than confirming it works as intended. Missing for 10: a documented, reliable full-data export mechanism (all chats, assets, code) confirmed to work post-cancellation, and independent verification that the promised fix was shipped.

   • [claimed-docs] “Export to PPTX, PDF, or HTML, share an org-scoped link, hand off to Claude Code, or send to the apps you already use daily.”
   • [community] “Dont use Claude Design, lost access to my projects after unsubscribing.”
   • [community] “It's pretty outrageous to lock out all your history just for canceling the subscription.”
   • [community] “If you export your data, all your Claude Design chats are in a design_chats directory along with the code, even if your account currently ha…”
   • [community] “Hi there, Thariq from the Claude team here. Sorry this is happening, we'll fix it ASAP. Claude's designs are HTML/CSS/JS that any editor can…”

   proof ↗⚑ flag

Plugins

1. designerExtend the editor from a large plugin community or marketplace — icons, data fillers, accessibility checkers, and more

   weight 2 · round to Penpot
   Penpot~partialcommunity6/10

   Docs confirm a plugin system and marketplace with 113+ plugins spanning accessibility, code, AI, and design-system tools, plus first-party docs on building custom extensions (penpot-docs-24, penpot-docs-5, penpot-docs-37). However, community feedback suggests the plugin ecosystem is still much smaller/younger than competitors' (penpot-comm-20 notes needing a 'bigger community' and missing pre-built element libraries), and there's no independent hands-on verification of marketplace breadth or quality. Missing for 10: independent reviews of specific popular plugins, marketplace UI/discovery evidence, and confirmation the 113+ figure reflects an actively curated, high-quality ecosystem rather than a small/early one.

   • [claimed-docs] “Plugins (113+): Community plugin marketplace with accessibility, code, AI, and design system tools.”
   • [claimed-docs] “Our plugin system brings the flexibility that devs need to create and publish their own custom extensions.”
   • [claimed-docs] “Use community plugins or build your own.”
   • [community] “Needs a bigger community. I know this project but still use Figma because in Figma I can simply use pre-created Elements from e.g. Ionic or …”

   proof ↗⚑ flag

   Claude Design—none0/10

   No evidence of a plugin marketplace or extensibility ecosystem for Claude Design (icons, data fillers, accessibility checkers, etc.); the only related mention is a single user connecting Claude Design to PenPot via MCP, which is a one-off integration, not a documented plugin/marketplace system.

   • [community] “Claude Design into PenPot via its MCP was a really neat flow, with correct prompts it even built out reusable PenPot components and design s…”

   proof ↗⚑ flag

Pricing

1. product managerPublished plan pricing makes seat types, tiers, and AI-feature gating clear before procurement

   weight 1 · round drawn
   Penpot—none0/10

   The evidence pack includes only scattered pricing-page fragments (unlimited teams, no file limits, community plugins) but no structured breakdown of seat types, plan tiers, or which AI features are gated behind specific plans. No pricing table, tier comparison, or AI-feature-gating disclosure is documented anywhere in the pack.

   • [claimed-docs] “Unlimited Teams - Collaborate freely with unlimited teams.”
   • [claimed-docs] “No file limits - Create as many as you need within your storage capacity.”
   • [claimed-docs] “No file limits. Create as many as you need within your storage capacity. ... Unlimited Teams. Collaborate freely with unlimited teams.”
   • [claimed-docs] “Use community plugins or build your own.”

   proof ↗⚑ flag

   Claude Design—none0/10

   The evidence pack shows product features, an enterprise admin toggle, and usage-bar mentions, but contains no documentation of seat types, pricing tiers, or explicit AI-feature gating tables for procurement decisions.

   ⚑ flag

2. designerThe free tier supports real design work — usable editor and collaboration without a paid seat

   weight 2 · round to Penpot
   Penpot~partialcommunity6/10

   Penpot's free tier documented as unlimited files and unlimited teams, and the core editor supports components, prototyping, and design tokens without gating behind a paid seat, which supports real collaborative design work. However, community reports raise concerns about performance/stability at scale (crashes, lag with many layers/pages) that could limit usability for complex real work, and there's no explicit pricing-page breakdown confirming free tier includes full collaboration features like comments/permissions. missing for 10: explicit free-tier feature/seat-limit breakdown from pricing page, independent confirmation that free tier collaboration (comments, real-time multiplayer) is unrestricted, and resolution of community-reported performance/stability issues affecting real design work.

   • [claimed-docs] “Unlimited Teams - Collaborate freely with unlimited teams.”
   • [claimed-docs] “No file limits - Create as many as you need within your storage capacity.”
   • [claimed-docs] “No file limits. Create as many as you need within your storage capacity. ... Unlimited Teams. Collaborate freely with unlimited teams.”
   • [claimed-docs] “Use community plugins or build your own.”
   • [claimed-docs] “A component is a layer or group of layers that can be reused multiple times across files. This can help you maintain consistency across a gr…”
   • [claimed-docs] “Penpot allows you to prototype interactions by connecting boards, which can act as screens.”
   • [community] “Unstable, very crash prone with just a few users designing 10 plus pages. And a huge memory hog too. I run it on Dedicated server with 64GB …”
   • [community] “I really wanted to like penpot, but when I tried a few months ago, simply navigating between pages (even on the example documents) was causi…”
   • [community] “Currently Penpot's performance starts dropping rapidly once you approach around 1000 layers. Most robust design systems I see run around 10-…”

   proof ↗⚑ flag

   Claude Design—none0/10

   The evidence pack contains no information about a free tier for Claude Design, its editor capabilities without payment, or collaboration limits for non-paying users. In fact, community reports indicate access is tied to an active subscription and is revoked upon cancellation (comm-11, comm-12, comm-13), which cuts against the idea of a functional free tier, but there's no direct claim or denial of a free tier's design capabilities to judge against.

   • [community] “Dont use Claude Design, lost access to my projects after unsubscribing.”
   • [community] “It's pretty outrageous to lock out all your history just for canceling the subscription.”
   • [community] “If you export your data, all your Claude Design chats are in a design_chats directory along with the code, even if your account currently ha…”

   proof ↗⚑ flag

1. ai-native userChoose where my data is stored (region/residency)

   weight 2 · round to Penpot
   Penpot✓fullclaimed7/10

   Penpot's self-hosting model directly delivers data residency choice: users can run Penpot in their own private cloud, on-premise, or any environment, giving full control over where data lives, and this extends to the MCP server which can also be self-deployed rather than routed through third-party clouds. Missing for 10: no first-party managed 'choose a region' feature for the SaaS-hosted offering, and no independent verification of self-hosting data-control claims beyond docs.

   • [claimed-docs] “Run it in your private cloud, on-premise, or any environment you control.”
   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”
   • [claimed-docs] “you can set up an instance of Penpot in just 3 minutes!”

   proof ↗⚑ flag

   Claude Design—none0/10

   No evidence in the pack mentions data residency, region selection, or geographic storage controls for Claude Design; only Enterprise admin enable/disable toggle is discussed, not residency.

   • [claimed-docs] “For Enterprise organizations, Claude Design is off by default. Admins can enable it in Organization settings.”

   proof ↗⚑ flag

2. ai-native userPrevent my data from being used to train AI models

   weight 3 · round to Penpot
   Penpot~partialclaimed6/10

   Penpot's self-host docs explicitly state 'Keep your data private. Penpot will never make business out of your data. Choose where and how to host it,' which directly addresses preventing data from being used for AI training by keeping it entirely under user control. However, this is a general privacy claim tied to self-hosting rather than an explicit AI-training opt-out policy or contractual clause, and there's no independent/third-party corroboration of this specific promise. missing for 10: explicit AI-training-specific policy/ToS language, independent verification of the no-training claim, and clarity on cloud-hosted (non-self-hosted) instance data usage.

   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”
   • [claimed-docs] “Run it in your private cloud, on-premise, or any environment you control.”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”

   proof ↗⚑ flag

   Claude Design—none0/10

   No evidence in this pack addresses training-data opt-out or privacy controls for Claude Design; the pack covers only product features, enterprise admin toggles, and unrelated community feedback.

   ⚑ flag

3. ai-native userControl data retention and deletion

   weight 2 · round to Penpot
   Penpot~partialclaimed4/10

   Penpot's self-hosting and 'keep your data private... decide where and how to host it' messaging (penpot-docs-28, penpot-docs-38) implies data control, and self-hosting inherently lets an org manage retention/deletion at the infrastructure level. However, there is no explicit documentation of data retention policies, file/version deletion mechanisms, or GDPR-style erasure APIs for AI-processed data. Missing for 10: explicit retention/deletion API or policy docs, confirmation of how MCP-accessed data is retained/purged, independent verification of deletion behavior.

   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”
   • [claimed-docs] “Run it in your private cloud, on-premise, or any environment you control.”
   • [claimed-docs] “Deploy the MCP server in your own environment without relying on third-party clouds.”

   proof ↗⚑ flag

   Claude Design—none0/10

   The evidence pack contains no documentation of user-controlled data retention or deletion settings for Claude Design; instead, community reports describe users being locked out of their design projects entirely after canceling a subscription, with Anthropic acknowledging the issue and promising a fix, rather than any dashboard/API for retention policy or deletion control.

   • [community] “Dont use Claude Design, lost access to my projects after unsubscribing.”
   • [community] “It's pretty outrageous to lock out all your history just for canceling the subscription.”
   • [community] “If you export your data, all your Claude Design chats are in a design_chats directory along with the code, even if your account currently ha…”
   • [community] “Hi there, Thariq from the Claude team here. Sorry this is happening, we'll fix it ASAP. Claude's designs are HTML/CSS/JS that any editor can…”

   proof ↗⚑ flag

4. ai-native userOpt out of telemetry and usage tracking

   weight 2 · round drawn
   Penpot—none0/10

   The evidence pack shows self-hosting and data-privacy marketing claims (e.g., 'Keep your data private,' 'never make business out of your data') but never mentions a specific telemetry/usage-tracking opt-out setting or configuration flag. Missing for 10: explicit documentation of a telemetry toggle or opt-out mechanism, confirmation of what usage data is collected, and any independent verification that self-hosted instances disable tracking by default.

   • [claimed-docs] “Keep your data private. Penpot will never make business out of your data. Choose where and how to host it, and decide who has access to it.”
   • [claimed-docs] “Penpot is the only open-source design platform that is truly deployment agnostic. Run it in your private cloud, on-premise, or any environme…”
   • [claimed-docs] “Run it in your private cloud, on-premise, or any environment you control.”

   proof ↗⚑ flag

   Claude Design—none0/10

   No evidence in the pack discusses telemetry/usage-tracking opt-out settings for Claude Design; the only related item confirms a separate usage bar, not a privacy control. Missing for 10: any documentation or setting for opting out of telemetry/usage tracking.

   ⚑ flag

1. designerPrototypes can carry advanced behavior — animation between states, conditions or variables, scroll effects — to feel like the real product

   weight 1 · round to Claude Design
   Penpot—none0/10

   Evidence only shows basic board-to-board click interactions (penpot-docs-12, penpot-docs-14); nothing in the pack mentions animation/transition effects between states, conditional logic, variables, or scroll-triggered effects in prototypes. This is an applicable axis for a prototyping tool, but no evidence supports these advanced behaviors.

   • [claimed-docs] “Penpot allows you to prototype interactions by connecting boards, which can act as screens.”
   • [claimed-docs] “Design responsive and rules-based interfaces with wireframes, interactive prototypes, and CSS Grid and Flex layouts.”

   proof ↗⚑ flag

   Claude Design~partialcommunity4/10

   Docs claim Claude Design produces 'interactive prototypes' that can be user-tested and 'look like the real product' via web capture, and community evidence confirms animated SVGs can be built, but with real friction (animations stripped by default sanitization, requiring workarounds). There is no evidence addressing prototype conditions/variables or scroll-based effects specifically. missing for 10: explicit support for state-based/conditional logic in prototypes, scroll-triggered effects, and clean first-party documentation of animation between screens/states without the SVG export bug.

   • [claimed-docs] “Turn a static mockup into a shareable, interactive prototype you can test with real people. No PRs, no code review.”
   • [claimed-docs] “Designers can turn static mockups into easily-shareable interactive prototypes to gather feedback and user-test, without code review or PRs.”
   • [claimed-docs] “You can also use the web capture tool to grab elements directly from your website so prototypes look like the real product.”
   • [community] “I've been using Claude Design to make animated SVGs. You have to ask it to first come up with a robust way to define the geometry and then a…”
   • [community] “Whatever tool prepares the SVG for export in Claude Design will strip animations as part of a sanitizing process; you have to export as a di…”

   proof ↗⚑ flag

2. designerWire screens into clickable prototypes with triggers, transitions, and overlays — no code required

   weight 2 · round to Penpot
   Penpot✓fullcommunity6/10

   Docs confirm board-to-board prototyping ("connect boards, which can act as screens") and marketing copy references "interactive prototypes"; a community user corroborates that "the prototype feature is working great." However, the evidence pack never details specific trigger types, transition/animation options, or overlay behavior, so the depth of the interaction model is unverified. Missing for 10: detailed documentation of trigger/transition/overlay options, hands-on walkthrough of prototyping fidelity, and independent reviews specifically praising or critiquing interaction complexity.

   • [claimed-docs] “Penpot allows you to prototype interactions by connecting boards, which can act as screens.”
   • [claimed-docs] “Design responsive and rules-based interfaces with wireframes, interactive prototypes, and CSS Grid and Flex layouts.”
   • [community] “It's really a fantastic tool, we are using it since several months in replacement of figma and Adobe XD. The prototype feature is working gr…”

   proof ↗⚑ flag

   Claude Design~partialcommunity4/10

   Docs confirm Claude Design turns static mockups into shareable, testable interactive prototypes without code (claude-design-docs-1, claude-design-docs-18), which covers the no-code prototyping goal, but there is no evidence of specific multi-screen wiring primitives like triggers, transitions, or overlays that a designer would use in traditional prototyping tools. Community feedback (claude-design-comm-6, claude-design-comm-9) discusses general design output quality but not this specific screen-linking mechanic. Missing for 10: explicit support for defining triggers/transitions/overlays between distinct screens, and any hands-on confirmation that multi-screen flows can be wired together.

   • [claimed-docs] “Turn a static mockup into a shareable, interactive prototype you can test with real people. No PRs, no code review.”
   • [claimed-docs] “Designers can turn static mockups into easily-shareable interactive prototypes to gather feedback and user-test, without code review or PRs.”
   • [community] “Used Claude Design to build the landing page for my side project over the weekend. Got something shippable in a few hours. It's great for 80…”
   • [community] “I tested it yesterday. Kinda impressive, but also design output is pretty boring.”

   proof ↗⚑ flag

Canvas

1. designerFrames can behave responsively — auto layout, constraints, and grids resize designs the way real UIs flex

   weight 2 · round to Penpot
   Penpot~partialclaimed5/10

   Penpot's marketing docs confirm CSS Grid and Flex layout support for responsive/rules-based interfaces (penpot-docs-14), but there is no detailed documentation of auto-layout resizing behavior, constraints on resize, or grid systems working the way native UI frameworks flex, and no independent/hands-on confirmation of these specific behaviors. Community evidence instead focuses on general performance/stability issues rather than confirming responsive layout fidelity. missing for 10: detailed docs on constraints-based resizing, auto-layout resize rules, hands-on validation that grids/flex truly mimic real UI responsiveness.

   • [claimed-docs] “Design responsive and rules-based interfaces with wireframes, interactive prototypes, and CSS Grid and Flex layouts.”
   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”

   proof ↗⚑ flag

   Claude Design—none0/10

   The evidence pack describes Claude Design as a prompt-driven generator that outputs HTML/CSS/JS mockups and prototypes, with only generic 'rich layout controls' to drag/resize/align elements (claude-design-docs-9) — there is no mention of Figma-style frames, auto layout, constraints, or grid systems that adapt responsively. Community feedback even notes the generated code is not maintainable/structured (claude-design-comm-15), suggesting it isn't built on a constraint-based vector layout engine. Missing for 10: any documentation of frames, auto-layout, constraint-based resizing, or grid systems enabling responsive behavior.

   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [community] “It makes absolutely beautiful designs, stunningly so, but the actual code is not something a human could ever maintain. Write-once, read-nev…”

   proof ↗⚑ flag

2. designerI get professional vector tools — pen and path editing, boolean operations, precise alignment — for real interface design work

   weight 3 · round drawn
   Penpot—none0/10

   The evidence pack covers AI/MCP integration, self-hosting, plugins, components, prototyping, and design tokens, but contains no documentation or hands-on mention of pen/path editing tools, boolean path operations, or precise alignment/snapping features. Community comments only compare general UI polish and performance to Figma without confirming these specific vector-editing capabilities.

   ⚑ flag

   Claude Design—none0/10

   Claude Design is a prompt-driven design/prototyping generator with only basic layout controls (drag, resize, align, sliders for spacing/color) — there is no evidence of pen/path vector editing, boolean path operations, or precision alignment tooling akin to professional vector editors. Community feedback even notes the underlying output is HTML/CSS/JS, not vector paths, and code is not maintainable, reinforcing it's not a vector design tool.

   • [claimed-docs] “use the adjustment sliders Claude creates to tune spacing and color. Use rich layout controls for quick visual and aesthetic shifts, includi…”
   • [community] “It makes absolutely beautiful designs, stunningly so, but the actual code is not something a human could ever maintain. Write-once, read-nev…”
   • [community] “I've been using Claude Design to make animated SVGs. You have to ask it to first come up with a robust way to define the geometry and then a…”
   • [community] “Whatever tool prepares the SVG for export in Claude Design will strip animations as part of a sanitizing process; you have to export as a di…”

   proof ↗⚑ flag

Files

1. designerExport production assets — PNG, SVG, PDF, and more — with per-layer presets, scales, and slices

   weight 2 · round to Claude Design
   Penpot—none0/10

   The evidence pack contains no first-party documentation describing PNG/SVG/PDF export options, per-layer export presets, scale factors, or slices — the closest data point is a community comment noting an inability to export a whole file (only a single layer) in past testing, which is negative rather than supportive. No docs, help-center pages, or hands-on confirmation of this specific export feature set are present.

   • [community] “Maybe I'm using it wrong... when I tried Penpot about a year ago, I had no way to export the diagram as one file like PDF or JPG. I could on…”

   proof ↗⚑ flag

   Claude Design~partialcommunity3/10

   Claude Design documents export to PPTX, PDF, and HTML, and community evidence confirms SVG export is possible (though it strips animations unless renamed), showing some baseline export capability exists. However, there is no evidence anywhere in the pack of PNG export, per-layer export presets, scale multipliers, or slice/asset regions — the core professional export controls the story asks about are entirely unevidenced. missing for 10: PNG export, per-layer export presets, scale export options, slice/asset export regions, any evidence of production-asset export workflows beyond whole-document PDF/PPTX/HTML.

   • [claimed-docs] “Export to PPTX, PDF, or HTML, share an org-scoped link, hand off to Claude Code, or send to the apps you already use daily.”
   • [community] “I've been using Claude Design to make animated SVGs. You have to ask it to first come up with a robust way to define the geometry and then a…”
   • [community] “Whatever tool prepares the SVG for export in Claude Design will strip animations as part of a sanitizing process; you have to export as a di…”

   proof ↗⚑ flag

2. developerThe file format is open or documented, so my designs stay portable and toolable outside the vendor's editor

   weight 1 · round to Penpot
   Penpot✓fullclaimed8/10

   Penpot explicitly documents an open file format built on open standards (SVG, CSS, HTML, JSON) and is open-source itself, explicitly contrasting with Figma's proprietary format, plus REST/webhook APIs and plugin system for external tooling. Missing for 10: independent third-party corroboration that .penpot files are actually opened/edited by external tools, and no dedicated file-format spec page (llms.txt/openapi probes 404).

   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON as well as Penpot's open file for…”
   • [claimed-docs] “Inspect and integrate code that actually matches design. Enjoy open standards like CSS, HTML, SVG and JSON”
   • [claimed-docs] “Penpot vs Figma: Open-source alternative with self-hosting, unlimited files, native design tokens.”
   • [claimed-docs] “Get extensive access to objects and properties to enable external rendering, workflow automation, enhanced app functionality, and more.”
   • [github] “developers feel at home using Penpot, because design is expressed as code, enabling a direct translation and shipping products faster.”

   proof ↗⚑ flag

   Claude Design~partialcommunity6/10

   Claude Design outputs are HTML/CSS/JS and can be exported to PPTX, PDF, or HTML (claude-design-docs-8), and an Anthropic team member explicitly confirmed 'Claude's designs are HTML/CSS/JS that any editor can handle' (claude-design-comm-14), supporting portability outside the vendor's editor. However, there was a real-world incident where users temporarily lost access to their design history/code after unsubscribing (claude-design-comm-11, claude-design-comm-12) before the fix, and community members note the generated code is hard for humans to maintain (claude-design-comm-15), undercutting practical toolability. missing for 10: a stable, always-available raw export/download path independent of subscription status, and clearer documentation of the underlying file format/spec beyond ad-hoc HTML/CSS/JS.

   • [claimed-docs] “Export to PPTX, PDF, or HTML, share an org-scoped link, hand off to Claude Code, or send to the apps you already use daily.”
   • [community] “Dont use Claude Design, lost access to my projects after unsubscribing.”
   • [community] “It's pretty outrageous to lock out all your history just for canceling the subscription.”
   • [community] “If you export your data, all your Claude Design chats are in a design_chats directory along with the code, even if your account currently ha…”
   • [community] “Hi there, Thariq from the Claude team here. Sorry this is happening, we'll fix it ASAP. Claude's designs are HTML/CSS/JS that any editor can…”
   • [community] “It makes absolutely beautiful designs, stunningly so, but the actual code is not something a human could ever maintain. Write-once, read-nev…”

   proof ↗⚑ flag

3. designerFull version history lets me name milestones, compare, and restore any earlier state of a file

   weight 2 · round drawn
   Penpot—none0/10

   No evidence in the pack mentions version history, milestone naming, diffing/comparison, or restore-to-earlier-state functionality; only component/plugin/self-host/AI features are documented. missing for 10: any mention of file version history, snapshot naming, diff/compare tools, or restore capability.

   ⚑ flag

   Claude Design—none0/10

   No evidence describes named milestones, version comparison, or restore-to-earlier-state functionality in Claude Design; community discussion only mentions losing access to 'history' after unsubscribing, not a versioning/milestone/compare/restore feature.

   • [community] “Dont use Claude Design, lost access to my projects after unsubscribing.”
   • [community] “It's pretty outrageous to lock out all your history just for canceling the subscription.”
   • [community] “If you export your data, all your Claude Design chats are in a design_chats directory along with the code, even if your account currently ha…”

   proof ↗⚑ flag