Labsco
MCP SERVER

Arcane MCP

by MikeCase

125 Docker management tools over Arcane, with a two-step confirmation handshake in front of anything destructive.

Containers, Kubernetes & Cluster Operations
Summary
The safety handshake is what makes 125 tools sane.

A model with `remove_container` and `prune_system` in reach is a bad afternoon waiting to happen; here those calls return a token and a warning instead of doing anything, and the agent has to stop and make a second call to proceed. Prune tools going dry-run by default and an audit log for every confirmed action are the same idea applied twice more.

What it is

A Python server that exposes the Arcane Docker management API to an agent: containers, images, volumes, networks, Compose projects, registries, vulnerability scanning, webhooks, the updater, and system management. Built with FastMCP.

What you get
  • 125 tools across 14 modules — Containers (18), Images (19), Volumes (18), Projects/Compose (17), Networks (9), System (9), Activities & Events (7), Registries (6), Vulnerabilities (6), Environments (5), Webhooks (5), Updater (3), Ports (1), Operations/Safety (2)
  • Every tool that removes, prunes, kills, restores or overwrites returns a `confirmation_token` instead of executing; the agent must call `confirm_operation` separately, and the token expires after 120 seconds
  • Prune tools — `prune_images`, `prune_volumes`, `prune_networks`, `prune_system`, `clear_activity_history` — default to `dry_run=True` and show what would be affected
  • Every confirmed destructive operation is written to a JSON-lines audit log, readable through `read_audit_log`
  • Vulnerability work is first-class: `scan_image_vulnerabilities`, `get_vulnerability_summary`, `list_all_vulnerabilities`, `ignore_vulnerability`, `get_scanner_status`
  • Resource tools take `env_id` (local Docker is "0", remote agents use UUIDs); remote agent calls can pass `agent_token`
Requirements

An Arcane instance and an API key. It is not on PyPI — install from source with `pip install git+https://github.com/MikeCase/arcane-mcp.git` or `uv tool install git+https://github.com/MikeCase/arcane-mcp.git`. Set `ARCANE_API_KEY` and `ARCANE_BASE_URL` in the environment or a `.env` file in the working directory. `ARCANE_MCP_AUDIT_LOG` moves the audit log off its default at `~/.arcane-mcp-audit.log`.

Setup effort

One command plus a key — pip install git+https://github.com/MikeCase/arcane-mcp.git, then supply credentials