CMA as an MCP server Guide
A thin MCP server that wraps the Claude Managed Agents Sessions API — so Claude Desktop **or** claude.ai web can start and chat with your org's hosted agents as if they were tools. `` User ─▶ Claude (Desktop or claude.ai) ─▶ MCP: send_message + wait_for_idle ─▶ CMA session ▲ │ └──────────────────────── agent's reply ◀─── stream-to-idle ◀──────────────────┘ ` Nine tools — eight are 1:1 with CMA endpoints, one (wait_for_idle`) is the SSE→request/response shim. Same handlers, two transports.
When to use CMA as an MCP server
A thin MCP server that wraps the Claude Managed Agents Sessions API — so Claude Desktop **or** claude.ai web can start and chat with your org's hosted agents as if they were tools. `` User ─▶ Claude (Desktop or claude.ai) ─▶ MCP: send_message + wait_for_idle ─▶ CMA session ▲ │ └──────────────────────── agent's reply ◀─── stream-to-idle ◀──────────────────┘ ` Nine tools — eight are 1:1 with CMA endpoints, one (wait_for_idle`) is the SSE→request/response shim. Same handlers, two transports.
How to use CMA as an MCP server
CMA as an MCP server is a single agent agent built on the Anthropic SDK framework. Set it up from the source repository, configure your model credentials, and invoke it for tasks that match its description. Review the safety profile below before running it against production data or systems.
Safety profile
Autonomy
Semi-autonomous
Sandbox-aware
Yes
Network access
Unspecified
Filesystem access
Unspecified