$ cd ..
$ mcp --connect claude▮
Expose the same task-board data to AI models as a small set of curated tools, over the same services/ layer as REST and GraphQL.
# steps
- 1Sign in & create an API key.
- 2Try a tool right here in the browser.
- 3Copy the connect snippet into Claude Code or Desktop.
// try a tool
// the wire · tool-call lifecycle
1
Claude picks tool
2
Arguments
3
services/
4
Structured result
// curated tools
// json-rpc 2.0 over the MCP transport
the exact messages exchanged with the model
// invoke a tool to see the JSON-RPC result envelope// connect for real
Claude Code · remote (HTTP)
claude mcp add --transport http protocol-lab http://localhost:8000/mcpClaude Desktop: Settings → Connectors → Add custom connector → URL http://localhost:8000/mcp
Local · stdio
claude mcp add protocol-lab \ -e MCP_API_KEY=pl_xxx \ -- uv run fastmcp run app/mcp_server.py
Because tools call the shared service layer, a task Claude creates shows up live on the real-time feed and fires webhooks.