Docker MCP

What QuantGeekDev/docker-mcp MCP server does

QuantGeekDev/docker-mcp MCP server connects an MCP client to Docker operations. Its documented scope covers creating standalone containers, deploying Docker Compose stacks, retrieving logs from a named container, and listing containers.

The server is intended for local Docker environments. The prerequisites list Docker Desktop or Docker Engine, Python 3.12 or newer, and UV. Claude Desktop is the documented client, with configuration examples for both production use and local development.

How it works

The MCP server exposes four tools:

• create-container creates a standalone container. Its documented input includes an image name, container name, port mappings, and environment data.
• deploy-compose deploys a Compose stack from a project name and Compose YAML content.
• get-logs retrieves logs for a specified container.
• list-containers returns a list of Docker containers and their status information.

Operations are issued by the connected MCP client, while Docker performs the underlying container actions. Compose deployments are supplied as YAML in the tool request rather than described as a repository path in the available examples.

Setup and configuration

The documented production configuration runs QuantGeekDev/docker-mcp MCP server with uvx:

{
  "mcpServers": {
    "docker-mcp": {
      "command": "uvx",
      "args": ["docker-mcp"]
    }
  }
}

Add this entry to the Claude Desktop configuration file. On macOS, the README identifies ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows, it identifies %APPDATA%/Claude/claude_desktop_config.json.

For development, the repository documents cloning the project, creating a Python virtual environment, activating it, and running uv sync. The development client configuration uses uv --directory <path-to-docker-mcp> run docker-mcp, so a local checkout path is required for that mode. Docker must also be installed and available to the host.

Tools and capabilities

The server covers basic container lifecycle and inspection workflows. You can create a container with selected image, name, port, and environment fields; deploy a Compose application; inspect logs for a particular container; and list containers for status monitoring.

These tools fit workflows such as starting a service from an image, bringing up a multi-service development stack, checking application output, or reviewing what is currently running.

Limitations and notes

QuantGeekDev/docker-mcp MCP server lists several unsupported or unavailable management areas: built-in environment-variable support for containers, volume management, network management, container health checks, restart policies, and container resource limits. Treat it as a focused interface for the documented creation, Compose deployment, log retrieval, and listing operations rather than a complete Docker administration layer.

The project is licensed under MIT. No API key, OAuth flow, or other external credential is documented; access depends on the local Docker installation and the MCP client's ability to launch the server.