Plumb (plumb-mcp)
⭐ If Plumb saves you tokens, star it on GitHub — it helps others find it.
The Figma → code MCP with a verification loop. Designs go in, normalised specs come out, and plumb-mcp verify drives headless Chrome to prove your rendered code actually matches what's in Figma.
📖 Full docs: https://tathagat22.github.io/plumb-mcp/ · 📦 npm: plumb-mcp · 🇨🇳 简体中文 · 🇯🇵 日本語 · 🇰🇷 한국어
Built for coding agents — Claude Code, Cursor, Windsurf, anything MCP-compatible. Reads Figma through a desktop-app plugin (no REST rate limits, works on every plan including Free), returns a compact normalised spec instead of the multi-hundred-thousand-token JSON the Figma API emits, and exports SVG icons + PNG images straight to disk.
How plumb is different
Three other Figma MCP servers worth knowing:
- Figma's official Dev Mode MCP — bidirectional, but plan-gated and metered.
- Framelink — thin REST wrapper. Two tools. No verification, inherits rate limits.
- cursor-talk-to-figma — bidirectional automation for designers working in Figma.
Plumb is the only one that closes the loop on code. plumb_verify tells you whether the code your agent shipped actually matches the design — colour-coded deltas, no pixel diff, runs in CI. And plumb_fit turns that into a self-healing loop: it scores the build 0–100, hands back the exact deltas to fix, and the agent iterates until it's pixel-perfect.
Run the loop three ways:
- In your editor —
plumb_fit(MCP). Claude Code / Cursor build, call it, and self-correct. Free; the agent is the generator. - From the terminal —
plumb-mcp fit <figma-url>generates a build, renders it headless in your Chrome, diffs, and corrects it pass-by-pass until it matches. (NeedsANTHROPIC_API_KEY.) - In the browser — the Playground: paste a Figma URL or pick a demo and watch it converge, with your own key. No install, no backend.
Are you hitting one of these?
If your agent landed here from an error, Plumb probably solves it.
| Error you're seeing | Why Plumb fixes it |
|---|---|
Figma Dev Mode MCP exceeded the 25k token cap · 351,378 tokens observed |
PDS dedups design tokens ($c1, $t1 …) and pre-resolves auto-layout to flexbox. A 178-node dialog comes back at ~2.6k tokens. |
Dev Mode MCP: 6 tool calls per month limit · Starter plan tool-call limit reached |
Plumb's plugin path has no per-call quota on any plan, including Free. |
Framelink figma-developer-mcp HTTP 429 · Figma REST API rate limit exceeded |
The plugin path doesn't touch REST. Zero rate limits. |
Variables API requires Enterprise plan · 403 Forbidden on variables |
Plumb reads Variables through the Figma Plugin API — works on every plan. |
Figma MCP returned 85% wrong layout · hallucinated structure |
Plumb returns structured PDS (not parsed prose) and ships plumb_verify + a plumb-mcp verify CLI that diff your rendered DOM against the design. |
Dev Mode MCP requires selection · "Open desktop app with the right selection" |
Plumb streams the whole file inventory the moment the plugin pairs. No per-call selection dance. |
Install: npm install -g plumb-mcp → plumb-mcp init.
Quick start
# 1. Install
npm install -g plumb-mcp
# 2. Wire into your editor — auto-detects Claude Code / Cursor / VS Code / Windsurf
plumb-mcp init
# 3. Sideload the Figma plugin (one-time). Find the manifest:
echo "$(npm root -g)/plumb-mcp/figma-plugin/manifest.json"
# Figma desktop → Plugins → Development → Import plugin from manifest…
# Run Plumb → click "Pair with Plumb" → done. Future runs collapse to a dot.
# 4. Optional — verify rendered code against Figma from the terminal
plumb-mcp verify http://localhost:5173/dashboard --url <figma-url>
# …or let it build + self-correct to pixel-perfect on its own (BYO Anthropic key)
plumb-mcp fit <figma-url>
Other install paths: npx plumb-mcp · docker run --rm -i ghcr.io/tathagat22/plumb-mcp:latest · build from source.
The fifteen MCP tools
| Tool | What it does |
|---|---|
plumb_status |
Self-description, key legend, connection state. Call first. |
plumb_outline |
Every screen in the file (id, name, size). |
plumb_node |
Extract a screen as compact PDS — by id or by name. |
plumb_query |
Pull a slice (skeleton / buttons / text / components) when a full screen would blow the token budget. |
plumb_describe |
Text-only visual description — per-region narrative + child summary, for image-blind harnesses. |
plumb_tokens |
Design-token table (colours, type, radii, shadows). |
plumb_selection |
The user's live Figma selection. |
plumb_assets |
Export icons (SVG) + images (PNG) — three modes: recursive, list (manifest only), or surgical by ids. |
plumb_screenshot |
Render any node to PNG/JPG. |
plumb_search |
Find nodes by name and/or type. |
plumb_components |
List components + instance usages. |
plumb_verify |
Diff your rendered layout against the design — structured deltas with ΔE2000 perceptual colour distance, shadow/rotation/flex-child/fill-stack checks, no pixel diff. |
plumb_fit |
The self-healing loop: plumb_verify plus a 0–100 convergence score and prioritised fixes, so the agent iterates to pixel-perfect instead of one-shot checking. |
plumb_fig_outline |
Headless: read a saved .fig file from disk and list every screen. No Figma desktop, no token. |
plumb_fig_node |
Headless: fetch one node from a saved .fig file by id. |
The agent's flow
plumb_outline() // 1. list screens
plumb_node({ name: "Settings" }) // 2. extract PDS
plumb_assets({ name: "Settings", ids: [...] }) // 3. pull only the icons you need
// 4. Build the UI — tag each element data-plumb-id="<el>" using the el from PDS
plumb_fit({ name: "Settings", rendered }) // 5. score it, fix the deltas, repeat
// 6. ...until plumb_fit returns done:true — pixel-perfect
plumb_fit is plumb_verify plus a convergence target — use it to iterate to a match. For a one-shot check, call plumb_verify instead. For the rendered payload shape, see the verify docs — or skip the in-browser capture entirely and use the plumb-mcp verify / plumb-mcp fit CLIs to drive headless Chrome end-to-end.
Two data paths
| Plugin (primary) | REST (secondary, headless) | |
|---|---|---|
| Rate-limited | No. Reads the in-memory document. | Yes. Free/Starter workspaces get very low budgets. |
| Token required | No. | Yes — FIGMA_TOKEN (figma.com → Settings → Security). |
| Variables | Yes, even on Free plans. | No — Variables REST API is Enterprise-only. |
| Headless / CI | No (needs Figma open). | Yes. |
Tools auto-pick the path. With the plugin paired, omit fileKey and pass
id or name. For the REST path, pass fileKey + id.
Architecture
Figma (desktop or browser, any plan)
│
│ Plumb plugin
│ • reads document + variables (Figma Plugin API, no rate limits)
│ • one-time "Pair with Plumb" click; collapses to a dot
▼
ws://127.0.0.1:31337 JSON control channel (paired, Origin-aware)
+ /upload/:key.:ext loopback HTTP for binary blobs — screenshots,
assets — POSTed straight to disk, no base64,
per-item ack for array uploads to keep Figma's
IPC from buffering and redelivering
▼
Plumb MCP server — `npx plumb-mcp` / `node dist/index.js`
│ • REST + plugin ingest
│ • Normalizer → Plumb Design Spec (PDS):
│ auto-layout → flexbox, tokens deduped, depth-stable `el` handles
│ (mints handles in a full pre-walk so the same node gets the same
│ el regardless of the requested depth — `plumb_verify` needs this)
│ • Version-keyed cache, fit-to-budget (maxTokens → auto-depth)
│ • Fifteen MCP tools (status / outline / node / query / describe /
│ tokens / selection / assets / screenshot / search / components /
│ verify / fit / fig_outline / fig_node)
▼
stdio MCP
▼
Claude Code · Cursor · VS Code · Windsurf
Configuration
.env (gitignored — never commit secrets):
FIGMA_TOKEN=figd_your_read_only_token # REST path only
PLUMB_FILE_KEY=… # for `npm run outline` etc.
PLUMB_NODE_ID=131:6950 # demo target
Cache and outputs:
- Cache —
~/.cache/plumb/v1/(TTL'd; override withPLUMB_CACHE_DIR). - Assets —
./plumb-assets/<screen>/(override withPLUMB_ASSETS_DIR). - Screenshots —
./plumb-screenshots/(override withPLUMB_SCREENSHOTS_DIR).
Testing
npm run typecheck # strict TS (server + plugin)
npm run build # bundle server + plugin
npm run smoke # MCP handshake; expects 15 tools
npm run check # offline fit-to-budget + cache verification
npm run bridge # simulated plugin + every tool offline
npm run prove # normalizer depth/token curve (fixture or live)
npm run outline # live: list every screen in your file (needs .env)
npm run connect # live end-to-end against a paired plugin
Layout
plumb-mcp/
├── src/
│ ├── index.ts # bin entry: stdio MCP server + bridge
│ ├── server.ts # registers the fifteen tools
│ ├── verify.ts # the plumb_verify comparison engine
│ ├── fit.ts # the convergence score + self-healing coaching
│ ├── cache.ts # on-disk version-keyed result cache
│ ├── assets.ts # writes exported assets to disk
│ ├── pds.ts # Plumb Design Spec types
│ ├── keylegend.ts # the compact-key legend (self-description)
│ ├── meta.ts # server name + version
│ ├── errors.ts # instruction-shaped error payloads
│ ├── figma/ # REST ingest + raw Figma types
│ ├── bridge/ # localhost WebSocket bridge to the plugin
│ ├── normalize/ # raw Figma → PDS (handles, layout, paint, …)
│ ├── tools/ # the fifteen MCP tools (one file each)
│ ├── fit/ # autonomous `plumb fit` loop — generate + serve
│ ├── render/ # shared headless / browser DOM capture
│ ├── cli/init.ts # `plumb init` — write editor MCP configs
│ ├── cli/fit.ts # `plumb fit` — self-healing build loop
│ └── util/ # round, estimateTokens, …
├── figma-plugin/
│ ├── manifest.json
│ ├── code.ts # main thread — reads, serializes, exports
│ └── ui.html # the panel (dot + pair button)
├── playground/ # the browser self-healing loop (Vite SPA, BYO-key)
├── scripts/ # smoke · check · bridge · connect · prove · outline
└── README.md # you are here
Security
- Loopback-only WebSocket bridge.
- Single paired plugin at a time; pairing is a deliberate click in the plugin
panel (one-time, then remembered via
figma.clientStorage). - Zero telemetry.
- No personal-access token needed for the plugin path; the REST path's token
is consumed only by the server's own outbound
fetchcalls.
License
MIT © Tathagat Maitray. See LICENSE.
