DesignDiff MCP
Verifies that AI-generated UI code matches Figma design specs by rendering components in a real browser, comparing computed CSS, and returning patch-ready fixes with scored parity reports.
README
DesignDiff MCP
AI implementation verification engine for Figma and MCP-compatible coding agents.
DesignDiff verifies that AI-generated interfaces match your Figma design system before they ship. It reads the Figma spec, renders your component in a real browser, compares computed CSS output, and returns patch-ready fixes — all inside your existing coding agent session.
Claude Code / Cursor / Windsurf
↓
DesignDiff MCP
↓
Figma REST API + Playwright renderer
↓
Scored diff + Patch-ready fix
↓
Agent applies → Re-checks: 94/100
Why
AI agents generate UI code from Figma specs and get it mostly right — but not exactly. The typical failure pattern:
- Token violations — hardcoded
#2563EBinstead ofvar(--brand-blue-600). Works today, breaks on the next brand refresh. - Spacing drift — padding is 12px not 16px because the agent approximated instead of reading the exact value.
- Missing states — hover, focus, and disabled were defined in Figma variants the agent didn't parse.
The component looks correct visually. That's what makes this dangerous. DesignDiff catches these errors before they land in a PR.
Quick start
1. Get your Figma API key Settings → Account → Personal access tokens
2. Add to your agent config
Claude Code (~/.config/claude/mcp.json):
{
"mcpServers": {
"designdiff": {
"command": "npx",
"args": ["-y", "designdiff-mcp"],
"env": { "FIGMA_API_KEY": "your-token-here" }
}
}
}
Same format for Cursor (.cursor/mcp.json), Windsurf, VS Code Copilot, and Zed.
3. Say to your agent
"Build the Button component from Figma file abc123, node 123:456,
running at localhost:6006 — and verify it matches the spec"
DesignDiff is called automatically when the agent generates or modifies a component.
Works best with Storybook, local preview routes, or any stable component URL your agent can render.
Tools
check_component_parity — ships now
Diffs the Figma design spec against computed browser CSS. Returns a scored parity report with ranked mismatches and a patch-ready fix when score drops below threshold.
check_component_parity({
file_id: "abc123", // Figma file ID from URL
node_id: "123:456", // Component node ID
component_url: "http://localhost:6006/story/button--primary",
code_path: "src/components/Button.tsx",
threshold: 80, // Score below which fix is generated (default: 80)
})
What it checks: spacing, color tokens, typography, borders, border-radius, interactive states.
What it returns: 0–100 parity score, ranked mismatches with consequences, pattern detection (e.g. "AI-generated code signature"), and a patch-ready git diff.
flag_stale_mappings — ships now
Cross-references Code Connect mappings in your repo against Figma version history and git blame. Surfaces components that have drifted silently, ranked by blast radius and days since divergence.
flag_stale_mappings({
file_id: "abc123",
repo_root: ".", // Path to your repo root
})
Reads from .figma/code-connect.json in your repo root.
generate_sync_patch — ships now
Generates a surgical patch for a specific component — token substitutions, prop corrections, missing state stubs. Output as git diff, JSX corrections, or CSS delta.
generate_sync_patch({
file_id: "abc123",
node_id: "123:456",
code_path: "src/components/Button.tsx",
format: "diff", // "diff" | "jsx" | "css"
})
audit_state_coverage — v0.2
Checks whether every interactive state defined in a Figma component set (hover, focus, disabled, error, loading) exists in code. Flags WCAG 2.4.7 risk for missing focus rings.
check_responsive_parity — v0.2
Renders at 375/768/1280/1920px and catches overflow, collapsed containers, and flex direction errors.
check_theme_parity — v0.2
Runs parity across Figma variable modes — light/dark, brand variants, high-contrast. Catches hardcoded colors that are invisible in light mode but break in dark.
Finding your Figma IDs
File ID: From the URL — figma.com/file/{FILE_ID}/...
Node ID: Right-click a component in Figma → Copy link → extract the node-id parameter. Format: 123:456. Note: the URL uses 123-456 (hyphen) but the API uses 123:456 (colon) — DesignDiff handles the encoding automatically.
How it works
-
Fetch — Calls the Figma REST API and extracts the full design spec: exact pixel values, token references, and every variant state from the component set.
-
Render — Launches a headless Playwright browser and renders your component at its real URL. Captures computed CSS — what the browser actually produces after cascade and runtime overrides resolve.
-
Score — Compares the two. Token violations score highest (they break the entire design system contract). Missing focus states are flagged with WCAG reference. Every mismatch includes the real-world consequence of not fixing it.
-
Fix — When score is below threshold, generates a patch-ready diff the agent applies in the same response, then re-verifies the score improved.
Configuration
| Environment variable | Required | Default | Description |
|---|---|---|---|
FIGMA_API_KEY |
✓ | — | Figma personal access token |
TRANSPORT |
— | stdio |
stdio or http |
PORT |
— | 3847 |
HTTP port (when TRANSPORT=http) |
HTTP transport
For use with non-stdio MCP clients:
FIGMA_API_KEY=xxx TRANSPORT=http npx designdiff-mcp
# → DesignDiff MCP v0.2 running at http://localhost:3847/mcp
Health check: GET /health
Development
git clone https://github.com/designdiff/designdiff-mcp
cd designdiff-mcp
npm install
npx playwright install chromium
npm run build
FIGMA_API_KEY=xxx npm start
Inspect tools interactively:
npm run inspector
This opens the MCP Inspector UI where you can call tools directly.
Roadmap
| Stage | What ships |
|---|---|
| v0.1 — Now | Rendered parity check · token compliance · patch-ready diffs |
| v0.2 | State coverage · responsive parity · theme parity · Storybook adapter |
| v1.0 | Unified quality score · CI gate · score history · Slack alerts |
Contributing
See CONTRIBUTING.md.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.