artokun/comfyui-mcp

The local-first, agent-native control plane for ComfyUI — MCP server + Claude Code plugin. 108 tools, 29 AI skills (Flux · WAN · LT2.3 · Qwen · Ideogram4 · Krea2). Author & run workflows, edit your live graph in natural language, manage models & custom nodes. Local, LAN, VPS, or Comfy Cloud.

What it solves

ComfyUI is a powerful node-based interface for generative AI, but it can be complex to master. This project provides a bridge that allows Large Language Models (LLMs) to act as an intelligent control plane, enabling users to drive ComfyUI through natural language instead of manual node manipulation.

How it works

It operates as an MCP (Model Context Protocol) server and includes a sidebar agent panel. It connects to various LLM providers—including Claude, ChatGPT, Gemini, or local models via Ollama—allowing them to interact with ComfyUI. The system doesn't just send prompts; it can author and edit the ComfyUI graph node-by-node, manage models, and execute complex workflows. It supports local installations, remote servers (like RunPod), and Comfy Cloud.

Who it’s for

  • ComfyUI users who want to use natural language to generate images, video, or audio.
  • AI Developers looking to integrate ComfyUI capabilities into agentic workflows using MCP.
  • Power users who want to automate complex, multi-step generative tasks through slash commands and autonomous agents.

Highlights

  • Multi-LLM Support: Works with Claude, ChatGPT, Gemini, and local models via Ollama or any OpenAI-compatible endpoint.
  • Agentic Control: Features specialized agents for exploring custom nodes, debugging workflow failures, and optimizing performance.
  • Deep Integration: Can edit the live ComfyUI graph, manage model downloads, and handle complex workflow compositions.
  • Versatile Deployment: Supports local, LAN, VPS, and cloud environments with a single configuration.