Controlling Docker Containers with AI via MCP

DevOps is complex. Remembering docker CLI arguments is hard. Let’s give Claude control of the Docker Daemon.

The Docker MCP Server

This server interfaces with the local Docker socket.

Capabilities

• list_containers
• inspect_container
• get_logs
• start/stop

Scenarios

1. Debugging a Crash

You: “The redis container keeps dying. Check the logs.” Claude: Calls docker_logs("redis"). Claude: “It looks like an OOM (Out of Memory) error. Exit code 137.”

2. Housekeeping

You: “Prune all stopped containers and unused images to free up space.” Claude: Calls docker_prune. “Deleted 4GB of unused images.”

3. Inspection

You: “What environment variables are set in the api-server container?” Claude: Calls docker_inspect("api-server"). “Here is the env list…”

Security Warning

Giving AI access to Docker is effectively giving it Root access (in many setups).

• Do not use this on a production server.
• Do use this on your local dev machine to speed up debugging.
• Always enable “Human Approval” for these tool calls.