Comfy MCP

About

Comfy MCP connects AI agents to a local ComfyUI environment through a server built on comfy-cli.

Key Features

• Run ComfyUI workflows from an AI assistant.
• Use the local installation’s models and workflow capabilities.

Use Cases

Add model inference or media generation tools to a developer-controlled agent.

Limitations

The local server and Comfy Cloud MCP have separate setup and compute requirements. A local installation does not imply access to hosted GPU resources.

Sources

• Project documentation: https://github.com/Comfy-Org/comfy-mcp
• Setup reference: https://support.comfy.org/articles/4321955727-agent-tools-mcp-comfy-cloud-mcp

MCP Setup

Prerequisites

• Python/pip; ComfyUI workspace; comfy-cli 1.14.0+.

Access requirements

• Partner API nodes need the corresponding Comfy API credentials; local diffusion depends on your own runtime and models.

Client setup

Claude Desktop

Transport: stdio

{
  "mcpServers": {
    "comfy-mcp": {
      "command": "comfy-mcp"
    }
  }
}

• Install: pip install comfy-mcp "comfy-cli>=1.14.0"
• Run comfy install if you need a workspace, then comfy launch and leave ComfyUI running.

Set COMFY_BIN if the MCP client cannot find the intended comfy executable.

Environment variables

• COMFY_BIN (optional path to comfy)
• COMFY_API_KEY (partner API nodes)

Setup docs: https://github.com/Comfy-Org/comfy-mcp