@amplify-ai/mcp-server
v2.0.1
Published
MCP server exposing the Amplify Canvas design system to AI agents (Pixel, Claude Code).
Readme
@amplify-ai/mcp-server
MCP server exposing the Amplify Canvas design system to AI agents (Pixel, Claude Code, etc.).
Tools
| Tool | Purpose |
|------|---------|
| list_components | All components in @amplify-ai/ui with one-line descriptions and tags. |
| get_props | Full prop signature for a component (variants, sizes, states, required props). |
| find_block | Search the templates package for blocks matching a use case (e.g. "checkout", "stepper"). |
| validate_usage | Validate a JSX snippet against component contracts — reports invalid props/values. |
| suggest_token | Given a raw value (color hex, spacing px), suggest the closest Canvas token. |
Transports
Two entry points:
amplify-mcp— stdio transport (used by Claude Code via.mcp.json).amplify-mcp-http— HTTP transport (used by Pixel runtime viaMCP_URL).
Local dev
npm install
npm run build
npm run dev # stdio
npm run dev:http # http on :3500Consuming from Claude Code
Add to ~/.claude/.mcp.json:
{
"mcpServers": {
"canvas": {
"command": "npx",
"args": ["-y", "@amplify-ai/mcp-server"]
}
}
}Consuming from Pixel
Pixel runs the HTTP transport as a sidecar (or in-process) and queries it via the standard MCP HTTP protocol — wired up in PR-P1.
