# Penpot MCP vs Figma MCP The Real Estate Card Test > [!question] Lingering Questions > - What rubric distinguishes "broken" from "clean, structured" layout output in the clip? None is given — the comparison is asserted, not shown against a visible artifact. > - Would the same identical-prompt test reproduce on a third MCP-enabled design tool, or is this a two-way framing that favors the vendor's narrative? > [!tip] Article Ideas > - none held — the source itself is too thin (44-second vendor promo) to develop into an article without independent reproduction first. > [!link] Cross Reference > - [[3. Resources/Permanent Notes/Agent-Readable Data Is Becoming the Default Plugin Integration Surface]] — related existing note (different source: WP Weekly) on the same "structure exposed to the agent sets the ceiling" mechanism, applied to WordPress plugins rather than design tools. > - [[3. Resources/Permanent Notes/Tool-Interface Choice Shapes Agent Architecture More Than Model Choice]] — related existing note on interface choice mattering more than model choice, from a CLI/MCP-basics source. > - [[3. Resources/MOCs/Visual Thinking MOC]] > [!quote] Vocabulary > - **design as code** (Penpot's term) — Penpot's underlying data model, where design files are structured/parseable rather than opaque canvases, which the clip credits as the reason its MCP produces agent-usable output. ## Bib capsule 44-second promotional YouTube clip from Penpot (open-source Figma alternative), demonstrating a controlled comparison ("the real estate card test") between Penpot's MCP server and Figma's MCP server on an identical complex prompt. Single vendor-produced test case; no shown artifacts or independent methodology. ## Author's thesis (what it's about / what's said in detail) Penpot runs the same complex design prompt through its own MCP server, Figma's MCP server, and unnamed competitors, and claims only Penpot produced structured, reusable, build-ready output while the others produced broken layouts and non-reusable elements. The clip credits this to Penpot's "design as code" approach — because the underlying file format is structured data rather than an opaque canvas, an AI agent gets richer context to both read and write design files, and that structural richness is what an agent needs to produce output a developer can actually build on. ## Key arguments (≤3) 1. **The comparison method is a controlled, identical-prompt test** — same requirements handed to multiple MCP servers, which at least in principle is reproducible, even though no artifacts are shown here. 2. **The causal claim is about data structure, not model quality** — Penpot attributes the gap to what its file format exposes to an agent, not to a smarter integration or a better prompt. 3. **This is vendor benchmark content and should be weighted accordingly** — no methodology, no visible output, no third-party reproduction; the underlying mechanism (structured data → better agent output) is plausible and worth testing independently, but the comparative claim itself is unverified. ## Claim atoms ### Identical-prompt test pits MCP servers on a card > [!quote] ID:rw_clip01 > Same requirements, same complex prompt. Figma and the other competitors produced broken layouts and elements you can't reuse. Where/why: ~00:10–00:18 of the clip — captures the controlled-comparison setup and the asserted competitor failure mode; the basis for any independent reproduction attempt. ### Design-as-code gives agents build-ready context > [!quote] ID:rw_clip02 > That's the real power of Penpot's design as code approach. AI agents get the rich context they need to read and write design files you can actually build on. Where/why: ~00:30–00:42, the clip's thesis line — states the mechanism Penpot credits for reusable, build-ready output; the claim worth testing independently. ## Permanent candidates - **Agent-readable design files beat opaque canvases for AI output** — new candidate from H2. No exact-source duplicate found; two related-but-distinct notes exist from other sources ([[3. Resources/Permanent Notes/Agent-Readable Data Is Becoming the Default Plugin Integration Surface]], [[3. Resources/Permanent Notes/Tool-Interface Choice Shapes Agent Architecture More Than Model Choice]]) and are linked rather than merged into. Promoted below. ## Processing checklist - [x] Checked for an existing Permanent note on this exact claim/source before drafting — none found; two related notes from other sources linked instead of treated as duplicates. - [ ] Reproduce the real-estate-card prompt across Penpot MCP, Figma MCP, and one more tool independently; capture artifacts before citing this comparison as evidence rather than vendor claim. - [x] Connected into [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] and [[3. Resources/MOCs/Visual Thinking MOC]]. ## Links - [[3. Resources/Readwise/Articles/Penpot MCP vs Figma McP The Real Estate Card Test]] — full saved source document - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — agent-tooling design choices belong to the vault's broader agent-architecture thread - [[3. Resources/MOCs/Visual Thinking MOC]] — design-file structure and how it's consumed is a visual-thinking-tooling question