arcane-mcp

An MCP server that exposes Arcane Docker management capabilities to AI agents via the Model Context Protocol. Built with FastMCP.

Covers the core Docker management API: containers, images, volumes, networks, Compose projects, registries, vulnerability scanning, port mappings, webhooks, system management, updater, activities/events, and environment management.

125 tools across 14 modules.

Install

Not on PyPI. Install directly from GitHub:

# pip
pip install git+https://github.com/MikeCase/arcane-mcp.git

# uv
uv tool install git+https://github.com/MikeCase/arcane-mcp.git

# or clone and install locally
git clone https://github.com/MikeCase/arcane-mcp.git
cd arcane-mcp
uv sync

Related MCP server: Docker MCP Server

Configuration

Set these environment variables (or put them in a .env file in the working directory):

ARCANE_API_KEY=arc_your_api_key_here
ARCANE_BASE_URL=http://localhost:3552

OpenCode

Add to your opencode.json:

{
  "mcp": {
    "arcane-docker": {
      "type": "local",
      "command": ["arcane-mcp"],
      "env": {
        "ARCANE_API_KEY": "arc_your_key_here",
        "ARCANE_BASE_URL": "http://localhost:3552"
      }
    }
  }
}

Or set environment variables globally and omit them from the config.

Tools (125 total)

Containers (18)

list_containers, inspect_container, create_container, start_container, stop_container, restart_container, kill_container, pause_container, unpause_container, remove_container, redeploy_container, commit_container, update_container, set_container_auto_update, exec_in_container, get_container_logs, get_container_stats, get_container_counts

Images (19)

list_images, inspect_image, pull_image, remove_image, tag_image, prune_images, get_image_counts, build_image, search_images, upload_image, get_image_history, get_image_export, get_image_attestations, scan_image_vulnerabilities, get_image_vulnerabilities, get_vulnerability_summary, check_image_update, check_all_image_updates, get_image_update_summary

Volumes (18)

list_volumes, create_volume, inspect_volume, remove_volume, prune_volumes, get_volume_counts, get_volume_sizes, get_volume_usage, list_volume_backups, create_volume_backup, restore_volume_backup, delete_volume_backup, download_volume_backup, browse_volume, read_volume_file, create_volume_directory, upload_to_volume, delete_volume_file

Projects / Compose (17)

list_projects, get_project, deploy_project, redeploy_project, remove_project, update_project, get_project_counts, project_down, restart_project, build_project, archive_project, unarchive_project, pull_project_images, get_project_compose, get_project_file, update_project_services, get_project_runtime

Networks (9)

list_networks, create_network, inspect_network, remove_network, connect_container_to_network, disconnect_container_from_network, prune_networks, get_network_counts, get_network_topology

Environments (5)

list_environments, get_environment, create_environment, update_environment, remove_environment

System (9)

get_docker_info, get_docker_version, prune_system, get_system_health, check_system_upgrade, trigger_upgrade, start_all_containers, start_stopped_containers, stop_all_containers

Activities & Events (7)

list_activities, get_activity, cancel_activity, clear_activity_history, list_events, get_environment_events, delete_event

Container Registries (6)

list_registries, create_registry, get_registry, update_registry, delete_registry, test_registry

Vulnerabilities (6)

get_vulnerability_summary_all, list_all_vulnerabilities, ignore_vulnerability, list_ignored_vulnerabilities, unignore_vulnerability, get_scanner_status

Webhooks (5)

trigger_webhook, list_webhooks, create_webhook, update_webhook, delete_webhook

Updater (3)

run_updater, get_updater_status, get_updater_history

Ports (1)

list_ports

Operations / Safety (2)

confirm_operation, read_audit_log

Usage

All resource tools accept env_id: str = "0" to target a specific Arcane environment (local Docker is "0", remote agents use UUIDs). Remote agent operations can pass agent_token: str | None for authentication via the X-Arcane-Agent-Token header.

Safety

Destructive operations use a three-layer safety system:

1. Confirmation tokens (two-step handshake)

Every tool that removes, prunes, kills, restores, or overwrites returns a short-lived confirmation_token instead of executing. The agent must call confirm_operation(token) as a separate tool call to proceed. Tokens expire after 120 seconds.

remove_container("my-app")
→ {"warning": "...", "confirmation_token": "a1b2c3d4e5f6", "target": "my-app"}

confirm_operation(token="a1b2c3d4e5f6")
→ {"success": true, ...}

This prevents an agent from blasting through a safety gate in one call — it must stop, process the warning, and make a second call.

2. Dry-run mode (prune/cleanup tools)

Prune operations (prune_images, prune_volumes, prune_networks, prune_system, clear_activity_history) default to dry_run=True, showing what would be affected without making changes:

prune_images()
→ {"dry_run": true, "warning": "Dry-run. Set dry_run=False to proceed.", "target": "all"}

prune_images(dry_run=False)
→ {"warning": "...", "confirmation_token": "..."}

3. Audit log

Every confirmed destructive operation is recorded to a structured JSON-lines audit log. Read it anytime:

read_audit_log(lines=10)
→ [{"timestamp": "2026-06-29T...", "action": "remove_container", "target": "my-app", "env_id": "0"}]

Set ARCANE_MCP_AUDIT_LOG environment variable to change the log file path (default: ~/.arcane-mcp-audit.log).

Development

git clone https://github.com/MikeCase/arcane-mcp.git
cd arcane-mcp
uv sync
uv run fastmcp run src/arcane_mcp/server.py

License

MIT