- 🎨 6 generation modalities — image, video, 3D, SVG, audio, and music
- 🔌 30+ providers under one interface
- 💻 Local server — runs on your machine, connects to the Comfy API
- 🔍 Built-in model search with BM25 text search
- 🧩 Drop-in compatible with Claude Desktop, Claude Code, and Amp
Requirements
- Node.js 20+
- pnpm 10
- Comfy API key (starts with
comfyui-) — get one from the Comfy partner dashboard
Installation
packages/mcp/dist/bin.js.
Configuration
Add the server to your MCP client config. The same JSON format works across Claude Desktop, Claude Code, and Amp:| Client | Config file |
|---|---|
| Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) / %APPDATA%/Claude/claude_desktop_config.json (Windows) |
| Claude Code | ~/.claude/mcp.json (or run claude mcp add) |
| Amp | Settings → MCP Servers, or ~/.config/amp/settings.json |
comfy-partner server.
To verify the server starts correctly before wiring it into a client, run it directly — it should print the JSON-RPC handshake and then idle:
Environment Variables
| Variable | Required | Purpose |
|---|---|---|
COMFY_API_KEY | yes | API key (comfyui-…). Sent as X-API-Key header. |
COMFY_API_BASE_URL | no | Override the proxy host (defaults to production Comfy API). |
COMFY_INLINE_LIMIT_KB | no | Cap (KB) for inlining image bytes in tool results. Default 600. |
COMFY_MCP_RESOURCE_LINKS | no | Force resource_link content blocks on/off. Defaults to an allowlist (claude-*, mcp-inspector). |
Available Tools
The MCP server exposes 10 tools. After a successful install, your MCP client should show all 10 registered under thecomfy-partner server.
| Tool | Description |
|---|---|
generate_image | Generate still images — text-to-image, edit, or upscale |
generate_video | Generate video clips — text-to-video, image-to-video, or lipsync |
generate_3d | Generate 3D meshes (GLB/FBX/OBJ) from text or a reference image |
generate_svg | Generate SVG vector illustrations from a text prompt |
generate_audio | Generate speech, voice clones, or sound effects from text |
generate_music | Generate instrumental or vocal music from a prompt |
media_upload | Upload a file (image/video/audio) and get a media:<id> reference for use in generate_* calls |
read_media | Fetch and render a media artifact inline — accepts an HTTPS URL or a media:… token |
models_explore | Discover models, search by capability, and read a model’s full call shape |
balance | Get the current account’s remaining balance in USD |
Example Prompts
After installing, try these in your AI assistant:Related
- Comfy Cloud MCP — Hosted MCP server for Comfy Cloud workflow execution
- Comfy Partner Nodes CLI — Use partner nodes from the command line