ComfyUI MCP Server
English
MCP Server for ComfyUI - Seamlessly integrate AI assistants (Claude Code, Claude Desktop, Gemini CLI) with ComfyUI through the Model Context Protocol.
Features
- 🎨 ComfyUI Integration: Connect AI assistants to ComfyUI via MCP protocol
- 🔧 Custom Node: Configure MCP parameters visually within ComfyUI
- 📦 Portable Setup: Use relative paths for easy environment migration
- 🚀 Auto-Start: MCP Server launches automatically when AI assistants need it
- ⚙️ Auto-Configuration: Automatically install and configure MCP for AI assistants
Quick Start
1. Start ComfyUI
Simply start ComfyUI with your normal startup script (no special MCP launcher needed):
```batch
Windows
run_nvidia_gpu.bat
Or any other ComfyUI startup script you normally use
```
2. Configure MCP in ComfyUI
Add the MCP Config Generator node to your workflow:
- Add `MCP Config Generator` node in ComfyUI
- Configure the node parameters:
- workflow_file: Select your workflow JSON file
- prompt_node_id: Node ID for text input (e.g., "45")
- output_node_id: Node ID for image output (e.g., "9")
- comfy_url: ComfyUI server URL (default: `http://127.0.0.1:8188`)
- auto_install: ✅ Enable (for first-time setup)
- auto_update_claude_code: ✅ Enable (if using Claude Code)
- auto_update_claude_desktop: ✅ Enable (if using Claude Desktop)
- auto_update_gemini_cli: ✅ Enable (if using Gemini CLI)
- Execute the workflow
The node will automatically:
- Install the MCP server module
- Generate configuration files
- Update AI assistant configuration files
3. Use with AI Assistants
That's it! The MCP server will automatically start when you use the AI assistant.
In Claude Code or Claude Desktop: ``` Generate an image of a cat in a sunny garden ```
The AI assistant will automatically:
- Start the MCP server (if not running)
- Call ComfyUI to generate the image
- Return the generated image
Manual Installation (Optional)
If auto-installation doesn't work, you can install manually:
```batch cd path/to/ComfyUI/custom_nodes/ComfyUI-MCP python -m pip install -r requirements.txt python -m pip install -e . ```
System Requirements
- Python >= 3.10
- ComfyUI
- Claude Code, Claude Desktop, or Gemini CLI
Architecture
This project uses a dual-layer architecture:
``` ComfyUI-MCP/ ├── src/comfy_mcp_server/ ← MCP Server core (auto-started by AI assistants) ├── nodes/mcp_config_node.py ← ComfyUI node (for configuration) ├── workflow/ ← Workflow JSON files ├── pyproject.toml ← Package configuration └── requirements.txt ← Dependencies ```
Troubleshooting
MCP server not working?
- Check that `auto_install` is enabled in the node
- Restart ComfyUI after first installation
- Verify the module is installed: `python -m pip list | grep comfy-mcp-server`
AI assistant can't find MCP server?
- Check that `auto_update_*` is enabled for your AI assistant
- Restart the AI assistant after configuration
- Manually check the configuration file:
- Claude Code: `~/.config/claude-code/mcp.json`
- Claude Desktop: `%APPDATA%/Claude/claude_desktop_config.json`
License
MIT
中文说明
ComfyUI MCP 服务器 - 通过模型上下文协议(MCP)无缝整合 AI 助手(Claude Code、Claude Desktop、Gemini CLI)与 ComfyUI。
功能特点
- 🎨 ComfyUI 整合: 通过 MCP 协议连接 AI 助手与 ComfyUI
- 🔧 自定义节点: 在 ComfyUI 中可视化配置 MCP 参数
- 📦 便携式设置: 使用相对路径,便于环境迁移
- 🚀 自动启动: AI 助手需要时自动启动 MCP 服务器
- ⚙️ 自动配置: 自动安装并配置 MCP 到 AI 助手
快速开始
1. 启动 ComfyUI
使用普通的启动脚本即可(不需要特殊的 MCP 启动器):
```batch
Windows
run_nvidia_gpu.bat
或者使用你平时使用的任何 ComfyUI 启动脚本
```
2. 在 ComfyUI 中配置 MCP
在工作流中添加 MCP Config Generator 节点:
- 在 ComfyUI 中添加 `MCP Config Generator` 节点
- 配置节点参数:
- workflow_file: 选择工作流 JSON 文件
- prompt_node_id: 文本输入节点 ID(例如 "45")
- output_node_id: 图像输出节点 ID(例如 "9")
- comfy_url: ComfyUI 服务器 URL(默认:`http://127.0.0.1:8188`)
- auto_install: ✅ 启用(首次设置时)
- auto_update_claude_code: ✅ 启用(如果使用 Claude Code)
- auto_update_claude_desktop: ✅ 启用(如果使用 Claude Desktop)
- auto_update_gemini_cli: ✅ 启用(如果使用 Gemini CLI)
- 执行工作流
节点会自动:
- 安装 MCP 服务器模块
- 生成配置文件
- 更新 AI 助手配置文件
3. 在 AI 助手中使用
完成!MCP 服务器会在你使用 AI 助手时自动启动。
在 Claude Code 或 Claude Desktop 中: ``` 生成一张阳光花园中的猫的图片 ```
AI 助手会自动:
- 启动 MCP 服务器(如果未运行)
- 调用 ComfyUI 生成图片
- 返回生成的图片
手动安装(可选)
如果自动安装不起作用,可以手动安装:
```batch cd ComfyUI所在路径/custom_nodes/ComfyUI-MCP python -m pip install -r requirements.txt python -m pip install -e . ```
系统要求
- Python >= 3.10
- ComfyUI
- Claude Code、Claude Desktop 或 Gemini CLI
架构说明
本项目采用双层架构:
``` ComfyUI-MCP/ ├── src/comfy_mcp_server/ ← MCP 服务器核心(由 AI 助手自动启动) ├── nodes/mcp_config_node.py ← ComfyUI 节点(用于配置) ├── workflow/ ← 工作流 JSON 文件 ├── pyproject.toml ← 包配置 └── requirements.txt ← 依赖项 ```
故障排除
MCP 服务器无法工作?
- 检查节点中是否启用了 `auto_install`
- 首次安装后重启 ComfyUI
- 验证模块已安装:`python -m pip list | grep comfy-mcp-server`
AI 助手找不到 MCP 服务器?
- 检查是否为你的 AI 助手启用了 `auto_update_*`
- 配置后重启 AI 助手
- 手动检查配置文件:
- Claude Code: `~/.config/claude-code/mcp.json`
- Claude Desktop: `%APPDATA%/Claude/claude_desktop_config.json`
许可证
MIT