Self-hosted Claude Managed Agents: run the whole agent service on your own servers (November 12, 2026)

What changed

November 12, 2026: Anthropic launched self-hosted Claude Managed Agents. Until now you could host only the sandboxes where an agent's tools run. Now you can run the whole Managed Agents service, the API, the agent loop and the storage for agents, sessions and session history, on your own cloud account or data center.

• Same API: your server has the same API as the hosted service, so the Anthropic SDK and the ant CLI work against it. You set ANTHROPIC_BASE_URL to your server and ANTHROPIC_API_KEY to your server's API key, not your Claude API key.
• Parts you run: the server (cma-server), a PostgreSQL 16 or later database (without one, data is kept in memory and a restart erases it), workers that run the agent's shell and file tools (ant beta:worker or an SDK worker), and an admin CLI (cma-server-admin).
• Three credentials: a server API key for your apps, an inference credential the server uses for model calls, and an environment key for your workers.
• Model provider: model calls go to the Claude API by default, or to Amazon Bedrock, Claude Models on Google Cloud, Microsoft Foundry or your own model gateway.
• Access: by request through your Anthropic account team. Only they can give you the release archive.

What your code may need to change, from Anthropic's differences page:

• Only self_hosted environments exist. A request for a cloud environment returns 400.
• The server does not serve the Messages, Models or Files APIs: /v1/messages and /v1/models return 404. Send those calls to the Claude API from a client that does not point at your server.
• No webhooks, scheduled deployments or dreams. No mcp_oauth vault credentials and no prebuilt docx, pdf, pptx and skills.