Containers your agent can run, inspect and clean up.
MCP Docker lets your agent do everyday container work for you: fetch and build images, start containers with the ports, folders and limits you want, read their logs and resource use, run a command inside one, copy files in and out, and manage the networks and volumes they use. It works with the Docker Engine you point it at.
What sets it apart from a shell is bookkeeping. It records containers, networks and volumes for each connection and can attempt their removal in one request. Review the result and remaining inventory before declaring cleanup complete. You choose approved registries for explicit pulls and pushes, and a container-count limit checked before creation. Set CPU, memory and storage budgets separately.
What you gain over typing container commands in a shell.
An agent that types container commands has to remember what it started, read text meant for people and clean up by hand. Clear results and resource records make those chores easier to review.
Clear, trimmed results
The agent reads the few details it needs instead of a wide table, which keeps the conversation short
- Typical alternative
- Reading the text a command prints
- Here
- Every answer is organised data, and long lists can be cut down to the few details you need
It cleans up after itself
Review removal counts and errors, then reconcile remaining resources; built and pulled images need separate retirement
- Typical alternative
- Containers, networks and volumes left behind by a failed experiment
- Here
- The service records what each connection created, and removes it on request, telling you about anything it could not remove
Safety rules you set
The limits are set in one place and apply to every connection
- Typical alternative
- An agent free to pull any image from anywhere and start any number of containers
- Here
- Choose approved pull/push registries and a container-count check before creation
Control only of what it owns
Deliberate adoption is required for a different container; verify its identity and coordinate with its existing controller first
- Typical alternative
- An agent that can stop or remove any container on the host
- Here
- The agent can start, stop and remove only the containers it created or deliberately took over
A whole workflow in a few requests
The agent gets the real outcome without scripting or waiting in a loop
- Typical alternative
- A shell script for build, run, wait, logs and copy
- Here
- The agent builds, starts, waits for the finish, reads the logs and copies files, each as a single request
One shape for any host you select
The agent uses the same requests whichever daemon you point it at
- Typical alternative
- A different command line for every host
- Here
- The same requests work against whichever supported Docker daemon you select
Ready in the image, one setting to point it.
MCP Docker comes in the MAIT image and updates with it (mait-image). Runs on Linux (amd64, arm64). On macOS and Windows, MCP Docker runs in the managed Linux environment that AIDC Launch provides. Your agent client starts it once you add one entry. It does not install or start Docker: it needs a running Docker Engine that the account launching it may use.
You choose the Docker host when you set it up. The local one on your machine is the default, and a host reached over a tcp:// address also works, from trusted clients only. Reach a remote host through a tunnel you set up yourself. Get started lists the settings.
You can also connect it through MCP Registry, so a single entry reaches all your tool servers.
Where it sits in the MAIT toolkit.
Where to read on.
- Get started, Use cases and Tools