Connecting a client

RyTask's MCP server speaks two transports. Both expose the same 49 tools and both authenticate with a personal access token (PAT).

The app shows you everything on this page, pre-filled for your server: open Settings → Agent access to see the HTTP endpoint, the stdio command, and short connect steps, next to the panel where you mint the token itself.

HTTP (streamable HTTP / SSE)

The remote transport lives on the API server at:

{API origin}/api/v1/mcp

Clients use POST (and GET for the SSE stream) with the token as a bearer credential:

Authorization: Bearer <your token>

The transport is stateless: there is no server-side session — every request is authenticated by its token, each request carries everything it needs. An invalid or revoked token gets an immediate JSON-RPC error (code -32001).

The public URL clients should use is whatever the MCP_PUBLIC_URL environment variable is set to on the server — that exact value is what the Agent access page displays. If it is unset, the page shows no HTTP endpoint (the integration is simply not advertised).

Claude Code

claude mcp add --transport http rytask https://rytask.example.com/api/v1/mcp \
  --header "Authorization: Bearer <your token>"

Replace the URL with the one shown on your Agent access page.

stdio (local)

For a client running on the same machine as RyTask, there is a local stdio transport. The command — exactly as the Agent access page shows it — is:

RYTASK_PAT=<your token> pnpm --filter @rytask/api mcp:stdio

Two practical notes:

• The script runs the compiled entrypoint (node dist/main.mcp.js), so the @rytask/api package must be built first (pnpm --filter @rytask/api build).
• It boots the full RyTask application context (it is a third entrypoint of the same backend, alongside the API and the worker), so it needs the same environment the API runs with — database, Redis, and so on.

Claude Desktop and other stdio clients

Add an mcpServers block to the client's configuration:

{
  "mcpServers": {
    "rytask": {
      "command": "pnpm",
      "args": ["--filter", "@rytask/api", "mcp:stdio"],
      "env": { "RYTASK_PAT": "<your token>" }
    }
  }
}

Run the client from (or set its working directory to) the RyTask repository so pnpm can resolve the workspace.

Once connected, the agent can list and call any of the 49 tools its token permits — see the MCP tools reference for the catalog and the security model for what "permits" means.