AI agents (MCP)

eDock is an MCP server. An AI agent in Claude Code, Claude Desktop, Cursor or VS Code can see your containers, read their logs, check ports and image scans, and run commands in throwaway containers.

Looking is free. Anything that changes something (start, stop, pull, run, exec) is asked in the agent's app first, and nothing happens if you say no.

Set it up

Engine › AI agents shows the exact command for this Mac. For Claude Code:

claude mcp add edock -- "/Applications/eDock.app/Contents/MacOS/edock-desktop" --mcp

For Claude Desktop, Cursor and VS Code, add this to their MCP settings (the page has a Copy button):

{
  "mcpServers": {
    "edock": {
      "command": "/Applications/eDock.app/Contents/MacOS/edock-desktop",
      "args": ["--mcp"]
    }
  }
}

The agent starts eDock with --mcp and talks to it over stdio. When the app is open, the agent works through it, and the app shows a note when an agent changes something. When it isn't, eDock runs without a window. Either way it uses the engine the app would: eDock's own, or the one picked in the toolbar.

What an agent can do

Tool		Asks first
`engine_status`	The engine, its version, and how many containers run
`containers_list`, `container_inspect`	Containers, with their state, ports and Compose project
`container_logs`	A container's recent log as text; `errors_only` keeps the lines that read like an error
`container_ports`	The ports a container really listens on, published or not
`domains_info`	Where containers answer (`<name>.edock.localhost`)
`http_requests`, `http_request`	Requests through container domains, while Requests records
`images_list`, `image_layers`	Images, and each Dockerfile line's size
`image_scans`, `image_scan_result`	Vulnerability scans
`volumes_list`, `networks_list`	Volumes and networks
`machines_list`	Linux machines
`container_files`, `container_file_preview`	A container's folders, and the start of a file
`recipes_list`	The services a recipe starts
`engine_log`	eDock's engine log, or only its warnings and errors
`k8s_status`, `k8s_items`	The Kubernetes cluster, its pods and services
`container_action`, `project_action`	Start, stop, restart, pause, unpause, kill or remove a container or a Compose project	Yes
`image_pull`	Pull an image	Yes
`project_up`	`docker compose up -d` (with `--build`) for a Compose project	Yes
`volume_clone`	Copy a volume into a new one	Yes
`recipe_start`	Start PostgreSQL, MySQL, ElyraSQL, SQL Anywhere, Redis, RabbitMQ… for development, or Askr for a project folder	Yes
`snapshots_list`	Snapshots of containers and volumes
`container_snapshot`, `volume_snapshot`	Take a snapshot	Yes
`container_snapshot_restore`, `volume_snapshot_restore`	Go back to one	Yes
`container_run`	Run a command once in a fresh container, then remove it	Yes
`container_exec`	Run a command in a running container	Yes

container_run is the safe place to try something: the container has no network unless the agent asks for it, stops after 60 seconds (at most 10 minutes), and is removed afterwards. It touches nothing on the Mac. Output longer than 64 KB is cut.

Agents can't create or remove volumes or networks, change the engine's settings, or reach Kubernetes beyond reading it.

Try it

Ask the agent things like:

Why does my api container keep restarting?
Which of my containers listen on a port that isn't published?
Does node:22-alpine have git? (it runs container_run and asks you first)
Which of my images have fixable critical vulnerabilities?

Check it from a terminal

The MCP Inspector lists eDock's tools and lets you call them by hand:

npx @modelcontextprotocol/inspector "/Applications/eDock.app/Contents/MacOS/edock-desktop" --mcp