Summary
Start and check — the smallest sensible surface for handing one agent system to another.
There is a temptation to expose a whole orchestration platform over MCP. Two tools is better: the crew's logic lives where it was built and tested, and the caller gets to trigger it and read the outcome. That keeps the failure surface tiny. The gap is documentation — no parameter schemas in the README means the shape of a kickoff call has to be discovered from the live server.
What it is
CrewAI's enterprise server, pointed at your own deployment rather than a shared address. The surface is deliberately two tools — this is a remote control for crews you have already built and deployed.
What you get
- kickoff_crew — start a deployed crew
- get_crew_status — inspect the run's status and results
Requirements
A CrewAI Enterprise deployment and a bearer token for it. Parameter schemas are not documented in the README.