Back to catalog
Pixelle MCP Server
An omnimodal AIGC framework that effortlessly converts ComfyUI workflows into MCP tools without writing code. Supports text, image, audio, and video generation using either local ComfyUI or RunningHub cloud mode.
An omnimodal AIGC framework that effortlessly converts ComfyUI workflows into MCP tools without writing code. Supports text, image, audio, and video generation using either local ComfyUI or RunningHub cloud mode.
Installation
UVX (temporary)
uvx pixelle@latest
Install via Pip
pip install -U pixelle
pixelle
From Source Code
git clone https://github.com/AIDC-AI/Pixelle-MCP.git
cd Pixelle-MCP
uv run pixelle
Docker
git clone https://github.com/AIDC-AI/Pixelle-MCP.git
cd Pixelle-MCP
cp .env.example .env
docker compose up -d
Features
- All-modality support: Supports full-modality conversion and TISV (text, images, audio/speech, video) generation
- Two execution modes: Local self-hosted ComfyUI environment + RunningHub ComfyUI cloud service
- ComfyUI ecosystem: Built on ComfyUI, inheriting all capabilities of the open ComfyUI ecosystem
- Code-free development: "Workflow as MCP tool" solution enabling code-free development and dynamic addition of new MCP tools
- MCP server: Based on MCP protocol, supports integration with any MCP client (Cursor, Claude Desktop, etc.)
- Web interface: Chainlit-based framework with UI controls and support for integrating other MCP servers
- One-click deployment: Supports installation via PyPI, CLI commands, Docker deployment methods
- Simplified configuration: Uses environment variable-based configuration schema
- Multi-LLM support: Supports OpenAI, Ollama, Gemini, DeepSeek, Claude, Qwen, and others
Environment Variables
Optional
PORT- Custom port for the service (default: 9004)
Usage Examples
Send any image to perform Gaussian blur processing through LLM
Generate images using text-to-image workflows
Automatically convert ComfyUI workflows into MCP tools
Notes
The system automatically launches a configuration wizard on first run to help you select the execution engine (ComfyUI/RunningHub) and configure the LLM. The web interface is available at http://localhost:9004, and the MCP endpoint is at http://localhost:9004/pixelle/mcp. Default login: username: dev, password: dev.