ComfyUI MCP Server

Build and run ComfyUI workflows with AI assistants. This MCP server connects Claude, Cursor, and other tools to your local ComfyUI install to list nodes, read schemas, validate workflows, and execute them (dry-run or full run).

No HTTP server required; it imports ComfyUI directly.

Features

• List all installed node types (including custom nodes)

• Fetch exact input/output schemas for any node

• Validate workflow JSON before running

• Execute workflows (reduced steps by default, optional full run)

Quick Start

Prerequisites

• Python 3.10+

• Local ComfyUI installation

• An MCP client (Claude Code, Claude Desktop, etc.)

Install

git clone https://github.com/zturner1/comfyui-mcp.git
cd comfyui-mcp
pip install -e .

Configure

# Windows (PowerShell)
$env:COMFYUI_ROOT = "D:\path\to\ComfyUI"

# Linux/macOS
export COMFYUI_ROOT="$HOME/ComfyUI"

Add to Claude Code

claude mcp add comfyui -- comfyui-mcp

Or add to Claude Desktop's config (claude_desktop_config.json):

{
  "mcpServers": {
    "comfyui": {
      "command": "comfyui-mcp",
      "env": {
        "COMFYUI_ROOT": "/path/to/ComfyUI"
      }
    }
  }
}

Usage Examples

List nodes:

"What upscaling nodes do I have installed?"

Validate then run:

"Validate this workflow JSON, then run it with 2 steps"

Full run:

"Run this workflow with full steps (steps = null)"

Configuration

AI Usage Guides

See CLAUDE.md and WORKFLOW_CHECKLIST.md for prompt guidance and validation rules.

License

MIT License - see LICENSE for details.