Connect Docker to Claude

Q: How far back do the logs go?

As far as Docker keeps them: with the default json-file driver everything since the container was created, unless you set max-size/max-file. A recreated container starts with an empty log. For longer history: Grafana with Loki.

You run the XenMCP Docker agent on your Docker host: a small open-source program (MIT) that can only read. It sees your containers, their state, logs, CPU and memory, and the events of the last days.

The agent runs behind your own reverse proxy (nginx, Caddy or Traefik) and answers only XenMCP. In XenMCP you add it in one step: the panel makes the key and shows the ready docker run command.

No Grafana or Loki needed: the agent reads the logs Docker itself keeps.

What you can ask Claude

Which containers are running, and which one crashed in the last 24 hours?
Why did api restart last night? Show me the errors before it.
Search the worker logs for “timeout” in the last hour.
Which container uses the most memory?
Is the health check of db green, and when did it last fail?

Typical tools

list_containers: Containers with image, state, uptime, ports and compose project
inspect_container: Exit code, OOM, restarts, health check, limits, volumes, networks – never environment variables
container_logs: Last lines, time windows, stderr only, text or regex search
container_stats: CPU, memory, network and processes right now
recent_events: Start, stop, crash with exit code, OOM, health check – up to 7 days
Start, stop, remove, exec: Not there – the agent sends Docker only reading requests

What you need to connect

A Docker host where you may start a container
A reverse proxy with HTTPS and an address for it, e.g. https://docker-agent.your-company.com/mcp
The command from the XenMCP panel (with its key)

Security

Read-only in the code: the agent sends Docker only a short list of GET requests and refuses everything else before it reaches Docker. The code is open and tested.
It never shows environment variables or start arguments – that is where passwords usually are.
It answers only with the right key and only XenMCP’s fixed IP address.
Hide containers with the label xenmcp.hide=true or a list of allowed names.
Logs can contain personal data. What a tool returns goes to your AI client – hide what is not needed.
XenMCP’s call log keeps only tool, status and duration – no logs.

Good to know

The agent is open source (MIT): github.com/xenMCP/docker-agent. Root on your host stays with you – XenMCP only holds a key to an agent that can read.

Built on

We use the open server XenMCP Docker agent (MIT), build the image ourselves from a pinned version and review the tool descriptions before release.

Frequently asked questions

How far back do the logs go?

As far as Docker keeps them: with the default json-file driver everything since the container was created, unless you set max-size/max-file. A recreated container starts with an empty log. For longer history: Grafana with Loki.

Can Claude stop or restart containers?

No. The agent has no such tool, and it does not send Docker anything outside its read list.

Do I need a domain?

Yes, an HTTPS address through your reverse proxy. Examples for nginx, Caddy and Traefik are in the agent’s guide.

Does it work with Docker Compose and Portainer?

Yes, the agent reads all containers on the host however they were started and shows their compose project.