MCP и вызов инструментов

comfyui-mcp

artokun/comfyui-mcp

A local-first, agent-native control plane for ComfyUI that provides an MCP server and sidebar agent to generate images, video, and audio, author and run workflows, and edit live graphs using natural language across any LLM.

★ 599Звёзды
⑂ 93Форки
44Открытые проблемы
TypeScriptЯзык
MITЛицензия
Q92Редакционная оценка

Скриншоты проекта

Скриншот comfyui-mcp Скриншот comfyui-mcp Скриншот comfyui-mcp Скриншот comfyui-mcp Скриншот comfyui-mcp Скриншот comfyui-mcp

Обзор

comfyui-mcp acts as a comprehensive control plane for ComfyUI rather than a simple prompt-to-image bridge. It allows AI agents to interactively build, modify, execute, and troubleshoot ComfyUI workflows node-by-node. The system ships with extensive model-specific expertise (samplers, CFG, resolutions, curated URLs) so connected LLMs can configure workflows accurately without trial and error. It supports local, LAN, VPS, and Comfy Cloud deployments, and can be accessed via a headless MCP server or an embedded ComfyUI sidebar panel.

Основные функции

  • 178 MCP tools for workflow execution, generation, and iteration
  • 36 AI skills providing model-specific generation guides and workflow recipes
  • 55 installer packs for one-command setups of specific model architectures
  • ComfyUI Agent Panel for live graph editing in the sidebar
  • Multi-provider support (Claude, ChatGPT, Gemini, Ollama, OpenRouter)
  • One-click remote tunneling via Cloudflare for hosted connectors
  • Autonomous agents for debugging, exploring nodes, and optimizing workflows
  • VRAM watchdog and cost guardrails for API credits

Требования, установка и быстрый старт

Add the MCP server to your Claude Code config (`~/.claude/settings.`):

{
"mcpServers": {
"comfyui": {
"command": "npx",
"args": ["-y", "comfyui-mcp"],
"env": {
"CIVITAI_API_TOKEN": ""
}
}
}
}

No need to clone the repository; npx downloads and runs it automatically. For the sidebar panel, search `comfyui-agent-panel` in ComfyUI-Manager.

Использование

With ComfyUI running and the MCP server configured, prompt your LLM (e.g., Claude) to generate an asset: `Generate an image of a sunset over mountains`. The agent will locate or download a checkpoint, build the workflow, execute it, and return the result. For remote connections, use `npx -y comfyui-mcp@latest connect https://<your-comfyui-url>`. To expose the server as a Streamable-HTTP endpoint, run `npx -y comfyui-mcp@latest --tunnel`.

Совместимость моделей и варианты использования

Compatible with any LLM that supports the Model Context Protocol or can run as an OpenAI-compatible endpoint. Explicitly supported backends include Claude (via Agent SDK), ChatGPT (via Codex app-server), Gemini, Ollama, LM Studio, llama.cpp, OpenRouter, DeepSeek, GLM, MiMo, and Kimi.

Лицензия и примечания о рисках

MIT License.

Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Релиз и сопровождение

Not stated in the repository metadata

MemPalace

mempalace/mempalace

★ 58KPython

litellm

berriai/litellm

★ 56,8KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython