
Comfy Pilot
MCP server + embedded terminal that lets Claude Code see, edit, and run your ComfyUI workflows.
Add to your client
Copy the config for your MCP client and paste it into its config file.
comfy node install comfy-pilotPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"comfy-pilot": {
"command": "python3",
"args": [
"/path/to/comfy-pilot/mcp_server.py"
]
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- ComfyUI
- Python 3.8+
- Claude Code CLI (installed automatically if not found)
About Comfy Pilot
Comfy Pilot is a ComfyUI custom node that bridges Claude Code and ComfyUI via an MCP server over stdio and an embedded terminal. It lets you describe workflows in natural language while Claude views the current graph, creates and connects nodes, runs the workflow, inspects generated images, manages custom nodes, and downloads models. The MCP server is configured automatically when ComfyUI starts.
Tools & capabilities (13)
get_workflowGet the current workflow from the browser.
summarize_workflowProduce a human-readable workflow summary.
get_node_typesSearch available node types with filtering.
get_node_infoGet detailed info about a specific node type.
get_statusGet queue status, system stats, and execution history.
runRun the workflow (optionally up to a specific node) or interrupt it.
edit_graphBatch create, delete, move, connect, and configure nodes.
view_imageView images from Preview Image / Save Image nodes.
search_custom_nodesSearch the ComfyUI Manager registry for custom nodes.
install_custom_nodeInstall a custom node from the registry.
uninstall_custom_nodeUninstall a custom node.
update_custom_nodeUpdate a custom node to the latest version.
download_modelDownload models from Hugging Face, CivitAI, or direct URLs.
What this server can do
Comfy Pilot provides tools for these capabilities — tap one to see every MCP server that does the same:
When to use it
- Build a complete workflow (e.g. an SDXL workflow with ControlNet) from a natural-language description.
- Inspect a generated preview image and have Claude adjust the workflow for more detail.
- Download a model (e.g. FLUX.1 schnell) and set up a workflow for it from scratch.
- Search for and install custom nodes from the ComfyUI Manager registry without leaving the chat.
- Run a workflow up to a specific node and review queue status and execution history.
Security notes
The plugin runs the Claude Code CLI as a PTY process inside ComfyUI and exposes tools that can edit workflows, install custom nodes, and download models to your machine. The MCP server runs locally over stdio with no authentication; only use it with a ComfyUI instance you trust. The download_model tool can fetch gated Hugging Face models using a token if provided.
Comfy Pilot FAQ
How do I install it?
Use the ComfyUI CLI with `comfy node install comfy-pilot`, install via ComfyUI Manager by searching for "Comfy Pilot", or git clone the repo into ComfyUI's custom_nodes folder. Restart ComfyUI afterward.
Do I need to configure the MCP server manually?
No. The plugin auto-configures MCP for Claude Code on startup. If it isn't connecting, you can manually add the comfyui server to ~/.claude.json with command python3 and args pointing to mcp_server.py.
The terminal says 'Command claude not found'. What do I do?
Install the Claude Code CLI. On macOS/Linux/WSL run `curl -fsSL https://claude.ai/install.sh | bash`; on Windows use the PowerShell or CMD installer from claude.ai.
Where do downloaded models go?
The download_model tool downloads models to your ComfyUI models folder and supports Hugging Face (including gated models with token auth), CivitAI, and direct download URLs.
Alternatives to Comfy Pilot
Compare all alternatives →Popular community server that feeds Figma layout data to coding agents via a Figma API token.
Generate beautiful, modern UI components from natural-language descriptions inside your IDE.
Give AI assistants full access to shadcn/ui v4 components, blocks, demos, and metadata across React, Svelte, Vue, and React Native.