pixelle-mcp-Image-generation

# 🙋‍♀️ Pixelle-MCP Frequently Asked Questions ### What's the difference between Pixelle MCP and traditional ComfyUI? - **Traditional ComfyUI**: Requires manual operations in the interface, workflows are relatively independent - **Pixelle MCP**: Intelligently invokes workflows through LLM, supports conversational operations, workflows can be automatically combined ### What installation methods are supported? Pixelle MCP provides three main installation methods: 1. **One-click Experience**: - Temporary run: `uvx pixelle@latest` - Persistent installation: `pip install -U pixelle` 2. **Local Development Deployment**: - Clone the source code and use `uv run pixelle` 3. **Docker Deployment**: - Use `docker compose up -d` (requires configuring .env file first) ### What is the default port? How to modify it? - **Default Port**: 9004 - **Modification Method**: Modify the `PORT` variable configuration in `.env` - **Access URLs**: - Web Interface: http://localhost:9004 - MCP Endpoint: http://localhost:9004/pixelle/mcp ### How to add custom MCP tools? 1. Create a workflow in ComfyUI 2. Set node titles according to DSL syntax specifications (e.g., `$image.image!:Input image`) 3. Export as API format file 4. Submit the workflow file in the Web interface and say: "Add Tool", LLM will automatically convert it to MCP tool 5. Refresh the page to use it ### Which output nodes are supported for workflow? The system automatically recognizes the following output nodes: - `SaveImage` - Image save node - `SaveVideo` - Video save node - `SaveAudio` - Audio save node - `VHS_SaveVideo` - VHS video save node - `VHS_SaveAudio` - VHS audio save node You can also manually mark: Use `$output.variable_name` in the node title ### What to do if workflow execution fails? 1. **Test in ComfyUI first**: Ensure the workflow runs normally in native ComfyUI 2. **Check parameter settings**: Confirm the parameter definition syntax in node titles is correct 3. **Check file paths**: Confirm input file paths are correct and files exist 4. **View execution logs**: Check detailed error information ### Which MCP clients are supported? Theoretically supports all clients that comply with the MCP protocol, including but not limited to: - Cursor - Claude Desktop - Other AI assistants that support MCP protocol ### How to configure multiple LLM providers? You can configure multiple LLM providers as alternatives in the configuration file `.env`, and the system will automatically select available services. ### How to batch import workflows? You can place multiple workflow files in the `data/custom_workflows/` directory, and the system will automatically load and convert them to MCP tools. Note: This method requires restarting the Pixelle-MCP service ### How to support LAN/external access? 1. Change `HOST` in `.env` to `0.0.0.0` 2. Change `PUBLIC_READ_URL` in `.env` to LAN/public address, such as: http://192.168.1.xx:9004 or http://www.xxx.com ### How to make the random seed change every time? - Set `seed` to `0` to randomize on each run. - Set `seed` to a positive integer (e.g., `123456`) to keep it fixed. ### How to Configure Pixelle MCP as a Standard MCP Server for Third-Party Applications? 1. First, deploy the Pixelle-MCP service according to the README instructions. 2. Then, configure the deployed service address in the third-party platform, following that platform's specific requirements. For example: To integrate Pixelle-MCP with Cursor, open the mcp.json file in Cursor and paste the following configuration, where http://localhost:9004/ should be replaced with your own Pixelle MCP server address: ```json { "mcpServers": { "pixelle-mcp": { "type": "streamable-http", "url": "http://localhost:9004/pixelle/mcp" } } } ``` --- 💡 **Tip**: If your question is not answered in this FAQ, feel free to join our community groups or submit issues on GitHub. We will continuously update this FAQ to help more users.