Arcane MCP Server
Manage your entire Docker infrastructure through natural language.
180 tools. One MCP server. Zero context switching.
Getting Started •
Tool Filtering •
What Can It Do •
Companion Skill •
All Tools •
Interactive Installer
Note: This is an independent, community-built project. It is not affiliated with, endorsed by, or maintained by the Arcane project or its maintainers. We build on top of Arcane's public API.
Coming from Portainer? Check out Portainer-Offload-Arcane — a guided migration wizard that moves your stacks, containers, volumes, registries, and users from Portainer CE/EE into Arcane. Run it once, then come back here to manage everything via natural language.
What is this?
Arcane MCP Server is a complete AI-powered Docker management bundle — an MCP server, a Claude Code plugin, and a companion skill, all in one package.
- The MCP server gives your AI assistant 180 tools to control your Arcane Docker platform — containers, images, stacks, Swarm clusters, security scans, and more. Filter that set down to a preset (see Tool Filtering) if all 180 is more than your context window can spare.
- The plugin wraps everything into a single install with guided configuration — no manual env vars or config files.
- The companion skill teaches your AI how to use those tools — safe deployment workflows, troubleshooting patterns, and guardrails so it doesn't accidentally nuke your volumes.
Instead of clicking through dashboards and running CLI commands, you just talk to your infrastructure:
"What's running on production?"
"Deploy this compose file to staging."
"Are any images vulnerable? Show me the critical ones."
"Scale the API service to 5 replicas."
Behind the scenes, the server translates your requests into the right API calls — with safety checks, retry logic, and formatted responses that actually make sense.
Why use it?
| Without | With Arcane MCP Server |
|---|
| Check container status | Open dashboard, navigate to environment, filter, scroll | "What's broken?" |
| Deploy a stack | Write compose, scp to server, ssh, docker compose up | "Deploy this compose file" |
| Find vulnerabilities | Configure Trivy, run scan, parse JSON output | "Run a security audit" |
| Update containers | Check each image, pull, recreate, verify | "Update everything" (dry-run first, obviously) |
| Clean up disk space | Prune images, volumes, networks one by one | "Clean up unused resources" |
Every tool carries safety annotations so your AI knows which operations are read-only and which ones need confirmation before running. No accidental docker system prune moments.
🆕 Recent update — v2.1: Tool filtering is here. All 180 tools in one turn eats your context window, so you can now pick a preset and ship only what you need. Details below.
Tool Filtering
Exposing all 180 tools to Claude every turn chews through your context window. Pick a preset to trim the active tool set — only the tools in that preset appear in tools/list:
| Preset | Scope | Tools |
|---|
commonly-used (recommended) | containers, images, projects, volumes, networks | ~52 |
read-only | every *_list / *_get / *_inspect / *_stats across all modules | ~60 |
minimal | dashboard + container list / get / counts | 5 |
deploy | projects, gitops, templates, registries, environments, build | ~40 |
full (default if never configured) | everything | 180 |
custom | your own module + per-tool picks | variable |
Configure interactively in Claude Code:
