Comfy MCP
The official Comfy Org MCP server: an agent runs workflows in your local ComfyUI and collects the finished images
- 311 ★ on the repository
- Last commit
Broad access
Why this level
- The tools run workflows with arbitrary nodes, manage the ComfyUI process and write files to disk
- Partner models and paid nodes spend Comfy credits on your account
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/comfy-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The tools run workflows with arbitrary nodes, manage the ComfyUI process and write files to disk
- Partner models and paid nodes spend Comfy credits on your account
This is third-party code. Review the repository files before installing.
What it does
Comfy MCP turns the ComfyUI you have installed into a toolset for an agent. The server starts as a stdio process, calls the comfy utility from comfy-cli on every request and hands its JSON response to the agent. By default everything targets the ComfyUI on your machine at `127.0.0.1:8188`. Among the 40 tools the main ones handle running and tracking: `run_workflow` executes a workflow JSON, `generate_image` makes a picture from text through a built-in template, `job` shows the state and cancels a task, and `fetch_outputs` collects the finished PNG files. The other tools search your installation for nodes, models and templates, custom nodes included. They also validate a graph, edit template slots and fan a workflow out into variants. You can manage the ComfyUI server itself too: start and stop it, read logs and stage input files. A separate `partner_generate` tool calls hosted partner models.
Who it is for. For people who run ComfyUI on their own computer and want to launch workflows, pick nodes and models and collect images straight from Claude Code or Cursor.
Good fit when
- You want the agent to run a saved workflow from `~/workflows/txt2img.json` and show the picture
- You want a quick image from a description without assembling a graph yourself: `generate_image` works
- You need to know which nodes and models are actually installed in your ComfyUI, custom ones included
- You need to run a batch of variants of one workflow and follow the queue: jobs are submitted asynchronously with wait, watch and cancel
- You need a ComfyUI on another machine you control: the address is set by the `COMFYUI_URL` variable
Not a fit when
- You have no GPU or installed ComfyUI: for cloud execution the project has a separate remote server, Comfy Cloud MCP
- You need guaranteed stability: the project is in beta and the toolset is still changing
- You want image generation without ComfyUI and node graphs: the server only works on top of it
Example request
Confirm my local ComfyUI is running, then run the workflow at ~/workflows/txt2img.json and show me the image.MCP
- Transport
- stdio
- Authentication
- not required
Environment variables
COMFY_BIN- Absolute path to the `comfy` command when it is not in the client environment.
COMFY_API_KEYsecret- Comfy key for partner nodes and models; only those need it.
COMFYUI_URL- Address of a ComfyUI on another machine instead of the local 127.0.0.1:8188.
Limitations
The server needs a separately installed comfy-cli of version 1.14.0 or newer: the comfy-mcp package does not pull it in and uses whichever comfy it finds on PATH or in the `COMFY_BIN` variable. Python 3.10 or newer is required. On macOS do not keep ComfyUI in Documents, Desktop or Downloads, otherwise system restrictions cause an Operation not permitted error. The `generate_image` tool works from a built-in template, and you can pass it your own checkpoint only after replacing the template. The project is in beta, under AGPL-3.0-or-later or a commercial license.
How to disable. Run `claude mcp remove comfy-mcp` or delete the `comfy-mcp` entry from the client configuration. The package is removed with `pip uninstall comfy-mcp`. ComfyUI itself and comfy-cli stay in place.
FAQ
How do I connect Comfy MCP to Claude Code?
Install with `pip install comfy-mcp "comfy-cli>=1.14.0"`, run `comfy launch` and execute `claude mcp add comfy-mcp -e COMFY_BIN=/path/to/venv/bin/comfy -- comfy-mcp`. Claude Desktop and Cursor use the same `mcpServers` block in their configuration.
How does Comfy MCP differ from Comfy Cloud MCP?
Comfy MCP runs as a subprocess on your computer and drives the local ComfyUI. Comfy Cloud MCP is a remote server at `https://cloud.comfy.org/mcp` that executes workflows on cloud GPUs and requires a Comfy Cloud account. You can install both.
Does the server spend money?
Local workflows are free. Only partner models and workflows with paid nodes spend credits, and the server asks for confirmation first. Such nodes need `COMFY_API_KEY`.
How do I work with ComfyUI on another machine?
Set the address in the `COMFYUI_URL` or `COMFYUI_HOST` variable: the run and job tools will go to the remote ComfyUI. Checkpoints must be installed on that machine.
Why is comfy-cli listed separately in the install command?
The comfy-mcp package does not declare comfy-cli as a dependency, so the server runs the comfy you chose. Without it the server starts, but every tool call reports that comfy is not found.
Discussion
No questions or reviews yet.
Related
MCP serversA lightweight Python MCP server: an agent generates and iteratively refines images, audio and video in a local ComfyUI
A local MCP server for editing and analyzing images: the agent crops frames, finds objects and reads text from photos
An MCP server for generating images and video from the agent: several models, a large prompt library and parallel tasks
Recraft's remote MCP server: an agent draws raster and vector images, edits a region, removes the background and upscales
ClipForge MCP
ClipForge
ClipForge MCP builds a vertical video from a topic or a product link: stock footage, Edge TTS voiceover and captions, no keys, composed with FFmpeg
An MCP server for Civitai: model and LoRA search, images with prompts, download commands, trend analysis, bilingual output
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Comfy MCP (this entry) | yes | not required | 311 | |
| MeiGen AI Design MCP | yes | API key | 1,793 | generates images and video with several cloud models and a prompt library |
| ima2-gen | no | 889 | a local generation studio in the browser, a Mac app or a CLI | |
| ComfyUI MCP Server | no | not required | 411 | a lightweight Python MCP server |
README in short
The reference covers all 40 tools with uniform argument naming. A separate section deals with partner nodes and the key they need.