MCP clients

Model Context Protocol is how editors and assistants connect to ClawQL. Hosts register search and execute for API work, and (when env is set) optional sandbox_exec, memory_ingest, memory_recall, ingest_external_knowledge, cache (CLAWQL_ENABLE_CACHE — ephemeral LRU, not the vault), and audit (CLAWQL_ENABLE_AUDIT — in-process trails). Cursor and Claude Desktop are common hosts.

What is MCP?

The MCP server runs as a subprocess (stdio) or over the network (HTTP). Your client configuration supplies environment variables such as spec paths or CLAWQL_PROVIDER. Single-spec execute uses in-process OpenAPI→GraphQL; no separate GraphQL service is required.

stdio (npm)

Use npx with -p clawql-mcp so the binary resolves from the package:

{
  "mcpServers": {
    "clawql": {
      "command": "npx",
      "args": ["-p", "clawql-mcp", "clawql-mcp"],
      "env": {
        "CLAWQL_SPEC_PATH": "/absolute/path/to/openapi.yaml"
      }
    }
  }
}

stdio (git checkout)

{
  "mcpServers": {
    "clawql": {
      "command": "node",
      "args": ["/absolute/path/to/ClawQL/dist/server.js"],
      "env": {
        "CLAWQL_SPEC_PATH": "/absolute/path/to/openapi.yaml"
      }
    }
  }
}

For development you can run tsx against src/server.ts instead; ensure the repo is built for production-style runs.

Streamable HTTP

For clawql-mcp-http, Docker, Kubernetes, or Cloud Run, configure your client with a url (not command). Copy .cursor/mcp.json.example to .cursor/mcp.json and set the base URL your deployment exposes. See Deployment. GraphQL debugging is available at /graphql on the same origin.