automaintainer/ docs
Dashboard →

MCP server

Drive Automaintainer from an agentic terminal (Claude Code, opencode, …) as native tools — no shelling out. am-cloud mcp runs a stdio MCP server backed by the same panel API and API key as am-cloud platform.

Configure your MCP client

Point your client at the am-cloud binary with the mcp subcommand. Most clients take a command + env:

{
  "mcpServers": {
    "automaintainer": {
      "command": "am-cloud",
      "args": ["mcp"],
      "env": {
        "AM_PANEL": "https://automaintainer.intrane.fr",
        "AM_API_KEY": "amk_..."
      }
    }
  }
}

Get AM_API_KEY from Settings → API key (or API key). --panel / --api-key flags also work if your client passes args instead of env.

Transport is stdio only (JSON-RPC over stdin/stdout) — there's no port to open and no inbound network exposure. initialize and tools/list work without credentials, so a client can introspect the tools before you set the key.

Tools

Tool What it does
am_status Account + connected worker status
am_repo_list List your repos (id, name, focus, enabled)
am_repo_run Run a repo and optionally wait for the verified result. Identify by repo_url (added if missing — the one-shot) or id; focus and wait optional. Returns {run_id, status, pr_url}
am_repo_set Configure a repo (focus, verify_cmd, objective, team_config, arch_context, auto_merge, issue_scout)
am_runs Recent runs (optionally by repo_id)
am_run_get One run by run_id (status, verify_status, pr_url)
am_probe Liveness-probe a cli/model before spending a run

Tool failures come back as MCP isError content (with the message), so the agent can read and react rather than crashing.

The one-shot, as a tool call

The headline flow — "enrich docs/ in this repo, run now" — is a single am_repo_run:

{ "name": "am_repo_run",
  "arguments": { "repo_url": "https://github.com/you/your-repo", "focus": "enrich docs/", "wait": true } }

It adds the repo if it's new, sets the focus, runs, and (with wait) returns the verified PR. Re-running the same repo_url reuses the repo (no duplicate).

CLI vs MCP

Both surfaces wrap the same API. Use the CLI for scripts and shell pipelines; use MCP when an agent should call AM as first-class tools. See the Agent cheatsheet for the full JSON contract and self-healing loops.