MCP Server Docker

What ckreiling/mcp-server-docker MCP server does

The ckreiling/mcp-server-docker MCP server lets an MCP-compatible client manage Docker resources through an agent interface. Its supported scope includes containers, images, networks, and volumes. It is suitable for local Docker experimentation, administration of a Docker engine, and conversational deployment workflows such as bringing up a WordPress and MySQL arrangement.

The included docker_compose prompt accepts a project name and a natural-language description of the desired containers. It guides the model through a plan-and-apply process, allowing the user to review a proposed plan or provide feedback before applying it. A later conversation can use the named project to inspect containers, volumes, and networks associated with that work.

How it works

The ckreiling/mcp-server-docker MCP server uses the Python Docker SDK's environment-based client setup. Local connections use the Docker environment available to the process, while a remote engine can be selected by setting DOCKER_HOST to an ssh:// host URL.

In addition to callable tools, the server provides resource templates for container logs and statistics. Clients can read docker://containers/{container_id}/logs as plain text and docker://containers/{container_id}/stats as JSON, substituting a container ID or name.

Setup and configuration

Claude Desktop can run the package through uvx using the mcp-server-docker package. The README identifies the Claude Desktop configuration paths for macOS and Windows. The server can also run inside a Docker container after building the repository image; that setup mounts /var/run/docker.sock so the server can reach the local Docker daemon.

For remote Docker access, add DOCKER_HOST to the MCP server definition with a value such as an SSH host URL. The Docker SDK documentation provides the remaining connection configuration details. No API key or hosted service is described in the supplied material.

Tools and capabilities

The ckreiling/mcp-server-docker MCP server exposes these container operations:

• List, create, run, recreate, start, stop, and remove containers
• Fetch container logs

Image operations include listing, pulling, pushing, building, and removing images. Network operations cover listing, creating, and removing networks. Volume operations cover listing, creating, and removing volumes.

The Docker Compose-style prompt is intended for conversational deployment and cleanup. Resource templates provide access to logs and runtime statistics without requiring the server to enumerate all running containers as resources.

Limitations and notes

Do not pass API keys, database passwords, or other sensitive values through containers configured by the model. The README warns that information exchanged with the LLM should be treated as compromised unless the LLM runs locally.

Docker is not a secure sandbox. Review containers and their requested configuration before applying changes because the server may affect the host through Docker access. For safety, the server does not support sensitive Docker options such as --privileged or --cap-add and --cap-drop.

The material documents Claude Desktop setup, but does not provide compatibility details for other MCP clients. The Docker-socket installation also grants the server access to the local Docker daemon, so it should be used only in an environment where that level of control is acceptable.