The problem
How it works
Comfy MCP is a local MCP server that wraps comfy-cli. You install it with comfy-cli, launch ComfyUI, and register the server in Claude Code with one command or in Claude Desktop or Cursor with a config entry. The agent then has 41 tools to generate an image from a text prompt, run a workflow JSON, follow and cancel jobs, fetch output files, search the nodes, models and templates your install actually has, validate and edit workflow graphs, turn videos into contact sheets it can inspect, and start or restart ComfyUI. A separate hosted Comfy Cloud MCP runs workflows on Comfy’s GPUs instead.
Results and limitations
The README says the server is in beta and that its core loop was checked end to end against a live local ComfyUI, from server info to a PNG on disk. That is a project test, not a reported creative project, so the case is marked planned.
Before you try it
Generation on your own GPU costs nothing, but partner-model tools and workflows that contain partner-API nodes spend Comfy credits; the server asks before those runs, though a custom node can still bill a service without a prompt, so check workflows you did not build. Installing nodes, updating ComfyUI or launching it on your network also asks for confirmation. If a local LLM shares the GPU, free its memory before large image jobs. Results depend on your setup and were not independently reproduced by Top 200 Repos.
Tools and repositories
- Tools
- Comfy MCP · comfy-cli · ComfyUI · Claude Code, Claude Desktop or Cursor
- Prerequisites
- Python 3.10 or later and a local ComfyUI install, ideally with a GPU.
Try it yourself: Make a consistent set of images
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
Using the ComfyUI tools, first check which checkpoints and templates my install has. Then generate [Number of images] images for [What they are for], in this style: [Style]. Use the same seed strategy and settings for all of them so they match, save them to [Output folder], and tell me the model, settings and seed used for each file. Do not install nodes or use any tool that spends credits.Look at every image before you publish it, and check the licence of the model you used.
Related videos
About Comfy-Org/ComfyUI
ComfyUI Tutorial for Beginners: Full Node Graph Basics (2026) (opens in a new tab)
The title names the project, the description links the project website.
Watch on YouTube (opens in a new tab)Independent creators. Listing a video implies no endorsement, sponsorship or partnership.
Sources
- Comfy-Org/comfy-mcp (opens in a new tab)
Project documentation · github.com · Undated · Source checked 12 Nov 2026, 18:05 UTC
What this source supports
- An MCP server built on comfy-cli to drive your own ComfyUI from Claude Code, Claude Desktop, Cursor or other MCP clients
- Tools to generate images, run workflows, monitor jobs, fetch outputs, preview media, search installed nodes, models and templates, and manage ComfyUI
- Status beta, 41 tools, core loop validated end to end against a live local ComfyUI
- Partner models and partner-API nodes spend Comfy credits and are confirmed first; custom nodes can call paid services without a gate
- Comfy Cloud MCP is a separate remote server that runs workflows on Comfy Cloud GPUs
