- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add artokun/comfyui-mcp --skill comfyui-node-registry -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add artokun/comfyui-mcp --skill comfyui-node-registry -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add artokun/comfyui-mcp --skill comfyui-node-registry -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add artokun/comfyui-mcp --skill comfyui-node-registry -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/artokun/comfyui-mcp/tree/HEAD/plugin/skills/comfyui-node-registry來自發布者
This skill covers writing a ComfyUI custom node pack and publishing it to the Comfy Registry (registry.comfy.org), the public catalog that powers ComfyUI-Manager. For using existing nodes in workflows, see the comfyui-core skill instead.
A node pack is a Python package placed under ComfyUI/custom_nodes/<name>/. The package __init__.py must export NODE_CLASS_MAPPINGS and NODE_DISPLAY_NAME_MAPPINGS; WEB_DIRECTORY is optional (only if the pack ships frontend JS).
ComfyUI/custom_nodes/my-node-pack/
├── __init__.py # exports the mappings ComfyUI scans for
├── nodes.py # node class definitions
├── pyproject.toml # registry metadata (required to publish)
├── requirements.txt # optional Python deps
├── .comfyignore # optional — exclude files from the published archive
├── LICENSE
├── README.md
└── web/js/ # optional frontend extension (see WEB_DIRECTORY)
__init__.pyfrom .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
# Optional: serve frontend JS/CSS from this folder (path relative to __init__.py)
WEB_DIRECTORY = "./web/js"
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"]
nodes.py)class ImageSelector:
CATEGORY = "example" # menu path where the node appears
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"images": ("IMAGE",),
"mode": (["brightest", "reddest", "greenest", "bluest"],),
},
"optional": {
"threshold": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01}),
"count": ("INT", {"default": 1, "min": 1, "max": 64}),
"label": ("STRING", {"default": "", "multiline": False}),
},
}
RETURN_TYPES = ("IMAGE",) # tuple of output data types
RETURN_NAMES = ("image",) # optional friendly output names
FUNCTION = "choose_image" # name of the method ComfyUI calls
OUTPUT_NODE = False # True for terminal nodes (e.g. SaveImage)
def choose_image(self, images, mode, threshold=0.5, count=1, label=""):
import torch
brightness = [torch.mean(img.flatten()).item() for img in images]
best = brightness.index(max(brightness))
return (images[best].unsqueeze(0),) # MUST return a tuple
NODE_CLASS_MAPPINGS = {
"ImageSelector": ImageSelector, # globally unique class_type key
}
NODE_DISPLAY_NAME_MAPPINGS = {
"ImageSelector": "Image Selector", # label shown in the UI
}
The options dict drives the UI widget for each input type:
("INT", {"default": 0, "min": 0, "max": 100, "step": 1})("FLOAT", {"default": 1.0, "min": 0.0, "max": 10.0, "step": 0.1})("STRING", {"default": "", "multiline": True})(["a", "b", "c"],), a literal list, becomes a dropdown("IMAGE",), ("LATENT",), ("MODEL",) etc. are typed connections (no widget)The executing method must return a tuple matching RETURN_TYPES, even for a single output (return (result,)).
pyproject.toml — Registry MetadataRequired to publish. comfy node init scaffolds this file. See references/pyproject.toml for a fully-commented example.
[project]
name = "my-node-pack" # unique & IMMUTABLE; lowercase, <100 chars, no "ComfyUI" prefix
description = "What this node pack does"
version = "1.0.0" # semantic version X.Y.Z — bump to publish a new version
license = { file = "LICENSE" } # or { text = "MIT License" }
requires-python = ">=3.10"
dependencies = [
"comfyui-frontend-package<=1.21.6", # optional — pin frontend if you ship UI
]
classifiers = [
"Operating System :: OS Independent",
]
[project.urls]
Repository = "https://github.com/you/my-node-pack" # REQUIRED — must be a valid repo URL
Documentation = "https://github.com/you/my-node-pack/wiki"
"Bug Tracker" = "https://github.com/you/my-node-pack/issues"
[tool.comfy]
PublisherId = "your-publisher-id" # the id after @ on your registry profile
DisplayName = "My Node Pack" # human-friendly name in the registry
Icon = "https://raw.githubusercontent.com/you/my-node-pack/main/icon.png" # square, max 400x400px
requires-comfyui = ">=1.0.0" # optional ComfyUI version constraint
.comfyignore uses .gitignore syntax; files listed are excluded from the published archive. Use it to drop tests, examples, large assets, and dev files.[tool.comfy].includes is the inverse: it force-includes folders that would otherwise be skipped (e.g. a bundled web/dist).@ on your profile page. It is globally unique and cannot be changed. Use it for PublisherId in pyproject.toml.Install comfy-cli (requires Python 3.10+; a virtualenv is recommended):
pip install comfy-cli
Versions are immutable. Once a version is published it cannot be modified or overwritten. To ship changes, bump version and publish again. To pull a bad version, deprecate it on the website (More Actions > Deprecate), which prompts users to upgrade rather than deleting it.
Automate publishing on every version bump. Add the API key as a repo secret named REGISTRY_ACCESS_TOKEN (Settings > Secrets and variables > Actions), then create .github/workflows/publish_action.yml:
name: Publish to Comfy registry
on:
workflow_dispatch:
push:
branches:
- main
paths:
- "pyproject.toml"
jobs:
publish-node:
name: Publish Custom Node to registry
runs-on: ubuntu-latest
steps:
- name: Check out code
uses: actions/checkout@v4
- name: Publish Custom Node
uses: Comfy-Org/publish-node-action@main
with:
personal_access_token: ${{ secrets.REGISTRY_ACCESS_TOKEN }}
main but only when pyproject.toml changes (i.e. when you bump version). workflow_dispatch allows manual runs.main, update the branches: list.pyproject.toml and publishes it, so the typical flow is bump version, commit, push to main, done.If your pack adds custom UI (widgets, sidebar tabs, menu items), set WEB_DIRECTORY in __init__.py and ship JS there. New frontend extensions should target the modern @comfyorg/extension-api rather than poking at legacy globals; pull in @comfyorg/comfyui-frontend-types for TypeScript types (npm install -D @comfyorg/comfyui-frontend-types). For the full frontend authoring workflow (defineExtension/defineNode/defineWidget and the defineSidebarTab/defineCommand/defineSetting shell APIs), see the sibling comfyui-frontend-extensions skill.
FUNCTION method must return (value,), not return value, even for one output.RETURN_TYPES without a comma. ("IMAGE") is a string, not a tuple. Write ("IMAGE",).INPUT_TYPES not a @classmethod. ComfyUI calls it on the class; missing the decorator breaks node loading.version instead; deprecate bad ones on the website.name after publishing. It's immutable and globally unique. Pick a good name (no "ComfyUI" prefix) up front.[project.urls].Repository. It's required; publishing fails without a valid repo URL.PublisherId. Use the id after the @ on your profile, not your display name.NODE_CLASS_MAPPINGS. The key is the class_type stored in workflow JSON. Changing it breaks every saved workflow that used the node.REGISTRY_ACCESS_TOKEN secret; never in pyproject.toml or the repo.comfy node publish.Authoring & publishing ComfyUI custom nodes to the Comfy Registry, covering node structure, pyproject.toml spec, comfy-cli publishing, and CI
---name: comfyui-node-registrydescription: Authoring & publishing ComfyUI custom nodes to the Comfy Registry, covering node structure, pyproject.toml…---# plugin/skills/comfyui-node-registry/SKILL.md常見問題
從發布者下載 plugin/skills/comfyui-node-registry/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/artokun/comfyui-mcp。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。
includes | [tool.comfy] | Array forcing extra folders into the published archive. |