Docker MCP — Container Management for AI Agents

What it is

Docker MCP is a Model Context Protocol server that gives AI agents direct access to Docker container operations. Through MCP tool calls, an agent can list running containers, start and stop services, inspect container logs, build images, and manage Docker networks and volumes.

This server is for developers who want their AI coding assistant to manage Docker environments without switching to a terminal. It is useful for debugging containerized applications, spinning up development databases, and automating container lifecycle tasks.

How it saves time or tokens

Instead of copying Docker commands between your terminal and AI chat, the agent executes Docker operations directly. It can inspect a failing container's logs, identify the issue, and suggest fixes in one conversation turn. The workflow provides the MCP configuration JSON for immediate setup.

How to use

1. Add the Docker MCP server to your .mcp.json:

{
  "mcpServers": {
    "docker": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-docker"]
    }
  }
}

1. Restart your MCP-compatible client (Claude Code, Cursor, etc.).

1. Ask the agent to manage containers:

'List all running containers'
'Show logs from the postgres container'
'Build and run the Dockerfile in this directory'
'Stop all containers and clean up unused images'

Example

{
  "mcpServers": {
    "docker": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-docker"]
    }
  }
}

With this config, the agent can run operations like:

# Agent executes these via MCP tool calls:
docker ps --format 'table {{.Names}}\t{{.Status}}\t{{.Ports}}'
docker logs --tail 50 my-api-server
docker build -t my-app:latest .
docker run -d --name my-app -p 3000:3000 my-app:latest

Related on TokRepo

• MCP Docker integration -- Full Docker MCP capability reference
• DevOps tools -- Infrastructure automation and container management

Common pitfalls

• The Docker daemon must be running on your machine. The MCP server connects to the local Docker socket and fails silently if Docker Desktop or dockerd is not started.
• Giving the agent Docker access means it can stop or remove containers. Be cautious in environments with production containers running alongside development ones.
• Building large images through the MCP server may time out with default settings. For complex builds, run them directly in the terminal.