Delegate tasks to a team of agents over MCP

Griffin exposes a Model Context Protocol (MCP) server. Your coding agent hands it a task in plain language, keeps working, and picks up the answer when it lands — while Griffin's agents do the work with their own tools and permissions.

Griffin on GitHub Quick start

Five task-shaped tools

griffin_send

Start a task in Persian or English. Returns the answer if it is quick, a question if Griffin needs input, or a task id.

griffin_wait

Long-poll a task for progress and the final answer, with SSE heartbeats so slow tasks survive proxies and CDNs.

griffin_reply

Answer a question the task asked you.

griffin_cancel

Stop a running task.

griffin_tasks

List your recent tasks and their states.

Connect Claude Code (streamable HTTP)

claude mcp add --transport http griffin https://griffin.example.com/mcp \
  --header "Authorization: Bearer grf_…"

Connect a stdio client (Claude Desktop, Cursor)

bin/griffin-mcp.mjs bridges stdio to the same endpoint and needs no npm install:

{ "mcpServers": { "griffin": { "command": "node",
  "args": ["/path/to/griffin/bin/griffin-mcp.mjs"],
  "env": { "GRIFFIN_URL": "https://griffin.example.com", "GRIFFIN_TOKEN": "grf_…" } } } }

Why delegate instead of giving your agent the tools

Each MCP client gets its own token and its own tool quota; credentials stay in Griffin and never reach the calling agent.
Irreversible actions go to a human for approval instead of being decided by whoever asked.
Tasks run for minutes without blocking the caller, and survive a server restart.
Every call is logged per client, so “it does not answer” is a lookup, not a guess.