The Agent panel driving ComfyUI end to end — reading what's installed locally, wiring the graph, freeing VRAM, and running the render
Quickstart
Add the server to Claude Code and generate your first image in ~2 minutes.
Tool Reference
38 tools, auto-documented from their live schemas.
Installation
Install ComfyUI and wire up the MCP server (local or remote).
Configuration
Environment variables, transports, tokens, and remote targeting.
What you can do
Generate
Text-to-image with
generate_image, plus ControlNet and IP-Adapter conditioning.Author workflows
Create from templates, modify, validate, visualize, and round-trip a compact DSL.
Manage models
Search HuggingFace/CivitAI, download, list, and remove models.
Custom nodes
Install, update, snapshot, and bisect node packs via ComfyUI-Manager.
Install & maintain
Install/update ComfyUI, inspect the environment, start/stop the server.
Local, remote, or cloud
Auto-detects a local install, points at a remote with
--comfyui-url, or talks to
Comfy Cloud via COMFYUI_API_KEY — no local ComfyUI needed.Works on macOS, Linux, and Windows, and auto-detects a local ComfyUI installation
and port. Source and issues live on GitHub.
Scope: local, remote, or Comfy Cloud
comfyui-mcp is the community MCP for local and remote ComfyUI — that’s the primary
target (Mac/Linux/Windows installs, RunPod, VPS, LAN, etc.).
For Comfy Cloud users, Comfy-Org ships official agent tooling — Comfy Cloud MCP (public beta) and the Comfy In-App Agent (private alpha) — maintained by the Comfy team and running on Comfy Cloud. If you have no GPU or want zero setup, use that (side-by-side). comfyui-mcp also includes a community cloud-mode (COMFYUI_API_KEY), so a single MCP can target all three deployment shapes from one config — pick whichever fits your workflow.