← all repositories
artokun/comfyui-mcp

An MCP Server That Turns Any LLM Into a ComfyUI Operator

It turns any LLM into a ComfyUI operator that edits live graphs, manages models, and runs workflows instead of just forwarding prompts.

comfyui-mcp
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

comfyui-mcp is an MCP server and sidebar agent that turns ComfyUI into something you drive with natural language. It generates images, video, and audio; authors and executes workflows; manages models and custom nodes; and edits your live node graph directly. It works with Claude, ChatGPT, Gemini, Ollama offline, or any hosted model through a single config, auto-detecting your local ComfyUI install or targeting a remote instance.

The interesting bit

The project ships with 37 model-specific “skills”—curated expertise on samplers, CFG values, resolutions, and download URLs for architectures like Flux and WAN—so the agent doesn’t guess. It also includes a built-in LLM Arena that scores different models on real ComfyUI tasks, plus hooks like a VRAM watchdog that warns before you OOM. That turns a dumb bridge into something closer to an embedded expert.

Key highlights

  • 115 MCP tools covering generation, workflow editing, model management, and custom-node installation.
  • 37 built-in skills with model-specific recipes and curated download URLs for Flux, WAN, LTX 2.3, and others.
  • Runs fully offline via Ollama, or on your existing Claude/ChatGPT/Gemini subscription with no extra API key.
  • Includes autonomous agents for debugging workflows, researching node packs, and optimizing performance.
  • One config targets local installs, LAN, VPS, or Comfy Cloud deployments.

Caveats

  • Comfy-Org already offers first-party agent tooling for Comfy Cloud users; if you lack a GPU or want zero setup, the official cloud path is currently the easier option.
  • OAuth and browser sign-in flows are planned but not yet implemented; remote auth currently relies on generated tokens.
  • The project is explicitly “local-first, not local-only,” which means cloud features exist but the primary focus is self-hosted installs.

Verdict

Worth a look if you run ComfyUI locally and want an LLM to actually operate the interface rather than just ping it. Skip it if you are all-in on Comfy Cloud and prefer first-party tooling.

Frequently asked

What is artokun/comfyui-mcp?
It turns any LLM into a ComfyUI operator that edits live graphs, manages models, and runs workflows instead of just forwarding prompts.
Is comfyui-mcp open source?
Yes — artokun/comfyui-mcp is open source, released under the MIT license.
What language is comfyui-mcp written in?
artokun/comfyui-mcp is primarily written in TypeScript.
How popular is comfyui-mcp?
artokun/comfyui-mcp has 500 stars on GitHub.
Where can I find comfyui-mcp?
artokun/comfyui-mcp is on GitHub at https://github.com/artokun/comfyui-mcp.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.