SkillFoxx

Comfy web UI MCP

A Gradio web UI over ComfyUI with MCP tools: an agent runs image, video, music and 3D generation through ready-made recipes.

  • 117 ★ on the repository
  • Last commit
  • Apache-2.0
MCP server

Extended access

Why this level

  • The service runs heavy GPU jobs and downloads models with Hugging Face and Civitai tokens kept in `yaml/config.yaml`.
  • Built-in protection is thin: without separate authentication an open port gives anyone on the network access to generation.
All reasons and checks

rioshiina47/comfy-webuiAuthor:RioShiina47

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add mcp/comfy-web-ui

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The service runs heavy GPU jobs and downloads models with Hugging Face and Civitai tokens kept in `yaml/config.yaml`.
  • Built-in protection is thin: without separate authentication an open port gives anyone on the network access to generation.

This is third-party code. Review the repository files before installing.

What it does

Comfy web UI is a modular Gradio web interface over ComfyUI. Workflow graphs are assembled on the fly from recipes, so an agent never wires nodes by hand: it describes the task and its parameters. The MCP part exposes high-level tools. The image module has eight of them, from listing tasks and models to `ImageGen_run_imagegen` and polling status by `task_id`. Separate tools run MiniMax-H3 and LTX-2.5 video, MiniMax Music 3 audio and image-to-3D through Pixal3D and TRELLIS.2 or Hunyuan3D-2. The same recipe works in the browser and for the agent.

Who it is for. Owners of a local ComfyUI who want to give an agent image, video, audio and 3D generation through ready-made MCP tools.

Good fit when

  • An agent picks a model and recommended samplers, then runs an image generation.
  • A single image becomes a textured 3D model.
  • A long video generation runs in the background and the agent polls its status by `task_id`.
  • A person and an agent share one set of recipes, in the browser and through MCP.

Not a fit when

  • ComfyUI is not running: the project connects to its backend and does not replace it.
  • The service would have to face the internet without hardening: the authors expect a trusted local network.
  • There is too little video memory for the chosen video and 3D models.

Example request

List the available image models, pick one suited to a portrait, generate a variant and report the task status.

MCP

Transport
http
Authentication
not required

Limitations

Python 3.10 or newer, a running ComfyUI (`http://127.0.0.1:8188` by default) and edits to `yaml/config.yaml`, such as the ComfyUI path, are required. Weight auto-download is off by default (`auto_download_models: false`) and models come from Hugging Face and Civitai. The README warns about limited built-in security and frequent changes to structure and API. The public MCP address in the README belongs to the author's demo space on Hugging Face.

How to disable. Stop `app.py` and remove the MCP entry from the client. Delete downloaded weights and cache in the ComfyUI folders by hand.

FAQ

Which tools does the image module have?

Eight: lists of tasks, architectures, models, features and samplers, model metadata, `ImageGen_run_imagegen` and `ImageGen_get_task_status` for polling.

How do I start the service?

Run `python app.py` in the project root. The interface opens at the address printed in the console, `http://127.0.0.1:7860` by default.

Does the project download weights itself?

Only if you enable `auto_download_models`. It is off by default to save bandwidth and disk.

Can the service face the internet?

Not without hardening. The authors suggest Tailscale or Zerotier, or a reverse proxy with authentication.

What does the video module do?

`VideoGen_H3_run` runs four MiniMax-H3 modes (`t2va`, `i2va`, `flf2va`, `ref2va`) and `VideoGen_LTX_2_5_run` runs five LTX-2.5 modes.

Discussion

No questions or reviews yet.

MCP servers

A lightweight Python MCP server: an agent generates and iteratively refines images, audio and video in a local ComfyUI

MCP serverExtended access413
Official

The official Comfy Org MCP server: an agent runs workflows in your local ComfyUI and collects the finished images

MCP serverBroad access341

A skill and MCP driver that let Claude Code, Codex and Gemini CLI run a local ComfyUI, with prompt recipes for 68 models

SkillExtended access121
Official

A Comfy Cloud plugin for Claude Code: commands to generate images, video, audio and 3D, connected through a hosted MCP server

PluginExtended access250
Official

A video editor with a local MCP server: an agent reviews the cut, edits the timeline and starts exports, ComfyUI generates

MCP serverExtended access531
Official

An MCP server for generating images and video from the agent: several models, a large prompt library and parallel tasks

MCP serverExtended access1.8K

README in short

The platform is split into modules with YAML descriptions of models, tasks and interface, and MCP functions register automatically from `*_mcp.py` files. The project talks to ComfyUI through dynamic graph assembly.

README badge

Report a problem