stitch-bridge

MCP server that gives AI coding agents the ability to design UIs through Google Stitch.

Your agent describes what it wants. Stitch generates production-ready HTML. stitch-bridge handles everything in between.

How it works

Your AI Agent  <-->  stitch-bridge (MCP)  <-->  Google Stitch SDK  -->  HTML / Screenshots

stitch-bridge runs as a local Model Context Protocol server over stdio. Any MCP-compatible client (Claude Code, Cursor, Windsurf, etc.) can connect to it and use 9 tools to create, edit, and manage UI designs, all through natural language prompts.

The Stitch SDK handles the actual generation using Gemini models. stitch-bridge fetches the results (HTML source, base64 screenshots) and returns them directly to the agent, so it can inspect, iterate, or save the output without leaving its workflow.

Related MCP server: Stitch-MCP

Quick start

1. Get a Stitch API key

Sign up at stitch.google.com and grab your API key.

2. Add to your MCP client

Add to your MCP config (e.g. .mcp.json, claude_desktop_config.json, or your editor's MCP settings):

{
  "mcpServers": {
    "stitch-bridge": {
      "command": "npx",
      "args": ["-y", "stitch-bridge"],
      "env": {
        "STITCH_API_KEY": "your-api-key"
      }
    }
  }
}

Or install globally:

npm install -g stitch-bridge

Then configure with command: "stitch-bridge" instead of using npx.

3. Use it

Ask your agent to design something:

"Create a project and generate a mobile login screen with email and Google sign-in"

The agent will call create_project, then generate_screen, and return the full HTML.

Tools

Options

Device types: MOBILE, DESKTOP, TABLET, AGNOSTIC

Models: GEMINI_3_PRO, GEMINI_3_FLASH (default)

Both are optional parameters on generate_screen, edit_screen, and generate_variants.

Output

Every tool returns JSON. The key outputs:

Generated screens return the full HTML source inline:

{
  "projectId": "proj-abc",
  "screenId": "scr-123",
  "html": "<!DOCTYPE html><html>..."
}

Screenshots return base64-encoded images:

{
  "screenId": "scr-123",
  "base64": "iVBORw0KGgo...",
  "mimeType": "image/png"
}

Multi-page sites return HTML per route:

{
  "projectId": "proj-abc",
  "pages": [
    { "route": "/", "screenId": "scr-1", "html": "..." },
    { "route": "/about", "screenId": "scr-2", "html": "..." }
  ]
}

Configuration

One of the two is required.

Development

npm install
npm run build      # Compile TypeScript
npm run dev        # Watch mode
npm test           # Run tests
npm start          # Run the MCP server

Requires Node.js >= 18.

License

MIT