MCP Server

@lerpa/mcp-server is a Model Context Protocolserver that exposes the Lerpa UI registry to AI agents. Point your editor's assistant at it and it can search the catalog and pull component source on its own — no copy-paste round trips.

What it provides

The server runs over stdio and exposes three tools:

list_components

List registry components, optionally filtered by category (ai, buttons, cards, forms, creative, feedback, navigation, tables, calendars, ecommerce, dashboard, auth, account, docs, blog).

get_component

Get the full registry item JSON for one component by id, including its embedded source code.

search_components

Fuzzy search across all items by name and description; returns up to 50 matches.

You can run it directly to confirm it starts:

Universal config

Most MCP clients share the same mcpServers shape. Drop this into your client's MCP config and you're done — the per-tool sections below are just where each client keeps that file.

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json in your project:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

Claude Code

Register it with the CLI:

Or commit a project-scoped .mcp.json at the repo root using the universal shape above so your whole team gets it.

Windsurf (Codeium)

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

VS Code (Copilot agent mode)

VS Code uses a servers key. Add .vscode/mcp.json in your project:

{
  "servers": {
    "lerpa": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

OpenAI Codex CLI

Codex configures MCP via TOML. Add to ~/.codex/config.toml:

[mcp_servers.lerpa]
command = "npx"
args = ["-y", "@lerpa/mcp-server"]

Google Antigravity

Open the in-app MCP settings and add a server using the universal mcpServers JSON shape:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

Trae (ByteDance)

Use the MCP marketplace / settings and add a custom server with the same mcpServers shape:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

Continue

Continue uses YAML. Add to ~/.continue/config.yaml:

mcpServers:
  - name: lerpa
    command: npx
    args:
      - "-y"
      - "@lerpa/mcp-server"

Cline

Add to Cline's cline_mcp_settings.json using the universal shape:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

Zed

Zed calls them context servers. Add to your settings.json:

{
  "context_servers": {
    "lerpa": {
      "command": {
        "path": "npx",
        "args": ["-y", "@lerpa/mcp-server"]
      }
    }
  }
}

Gemini CLI

Add to ~/.gemini/settings.json using the universal mcpServers shape:

{
  "mcpServers": {
    "lerpa": {
      "command": "npx",
      "args": ["-y", "@lerpa/mcp-server"]
    }
  }
}

JetBrains AI & Warp

Both expose an in-app MCP settings panel. Add a new server with the universal mcpServers JSON shape shown above.

Running locally (before npm publish)

Until @lerpa/mcp-server is published to npm, run the built server by absolute path instead of npx. Build it first with pnpm --filter @lerpa/mcp-server build, then point your client at dist/index.js:

{
  "mcpServers": {
    "lerpa": {
      "command": "node",
      "args": ["/abs/path/packages/mcp-server/dist/index.js"]
    }
  }
}

The MCP server is backed by the same registry as the CLI, so anything an agent finds, it can install. For a guided overview, see the tutorial.