AI-powered design generation in Figma — 100% design system compliant.
Bridge turns Claude Code into a designer that knows your design system inside out. It extracts, documents, and uses your real Figma components, tokens, and text styles to generate production-ready designs.
You describe what you want
→ Claude consults the knowledge base (your DS, documented)
→ Claude writes the spec (exact components, tokens, layout)
→ Claude generates in Figma via figma_execute (real DS components, bound variables)
→ You review in Figma
Bridge is two things:
- A CLI (
bridge-ds init) that scaffolds your project with the design workflow skill - A Claude Code skill (
/design-workflow) that handles the intelligence — spec writing, DS knowledge, Figma generation
The transport layer is figma-console-mcp, an MCP server that gives Claude native access to Figma (57+ tools).
Claude Code ──MCP──> figma-console-mcp ──WebSocket──> Figma Desktop
(your DS library,
real components,
bound variables)
- Claude Code installed
- Node.js 18+
- Figma Desktop (not the web app)
- A Figma file with a published design system library
claude mcp add figma-console -s user \
-e FIGMA_ACCESS_TOKEN=figd_YOUR_TOKEN \
-- npx -y figma-console-mcp@latestGet your token from Figma Settings → Personal access tokens.
- Run
npx figma-console-mcp@latest --print-pathto find the plugin directory - In Figma Desktop: Plugins → Development → Import plugin from manifest...
- Select
figma-desktop-bridge/manifest.json - Run the plugin in your Figma file
cd your-project
npx @noemuch/bridge-ds initThis scaffolds:
.claude/skills/design-workflow/— the workflow skill + references.claude/commands/design-workflow.md— the/design-workflowslash commandspecs/— directory for active, shipped, and dropped specs
Open Claude Code in your project:
/design-workflow setup
Claude will:
- Extract your entire DS from Figma (
figma_get_design_system_kit) - Analyze every component, token, and style
- Generate intelligent guides (when to use what, decision trees, pattern catalog)
- Ask for product screenshots to document layout patterns
/design-workflow spec a settings page for account information
Claude consults the knowledge base, identifies the right pattern, components, and tokens, and writes a complete spec. Then:
/design-workflow design # Generate in Figma (atomic, verified)
/design-workflow review # Validate against spec
/design-workflow done # Archive and ship
setup (once) → spec → design → review ──→ done
↑ | |
└── iterate ─────────┘ learn ←┘
(diff corrections,
extract preferences)
No design without a validated specification. Claude knows exactly which components, tokens, and layout patterns to use because it has your DS documented.
Designs are generated in 4-6 small sequential scripts (~30-80 lines each). After each step, Claude takes a screenshot and verifies before continuing. Bug in step 3? Fix and re-run step 3 only.
Zero hardcoded hex colors. Zero recreated components. Everything imported from your library via importComponentByKeyAsync, bound to variables via setBoundVariableForPaint.
Blocking checks at every phase transition: spec validation, pattern matching, pre-script element audit, visual fidelity review, DS component reuse audit.
The knowledge base is what makes Bridge different from "just executing Figma scripts". During setup, Claude builds a complete understanding of your DS:
knowledge-base/
registries/ ← Raw DS data (components, variables, text styles)
guides/
design-patterns.md ← Layout patterns from your product screenshots
tokens/ ← When to use which color, spacing, typography
components/ ← Decision tree: "I need X" → use component Y
patterns/ ← Form, navigation, feedback, multi-step patterns
assets/ ← Icons, logos, illustrations catalog
ui-references/ ← Product screenshots for pattern extraction
| Command | What it does |
|---|---|
/design-workflow setup |
Extract DS + build knowledge base |
/design-workflow spec {name} |
Write a component or screen spec |
/design-workflow design |
Generate in Figma from active spec |
/design-workflow review |
Validate design against spec + tokens |
/design-workflow done |
Archive spec and ship |
/design-workflow drop |
Abandon with preserved learnings |
/design-workflow learn |
Diff corrections, extract learnings |
/design-workflow sync |
Incremental DS sync (no full re-setup) |
/design-workflow status |
Show current state, suggest next action |
MIT