Skip to main content
ComfyUI provides MCP servers and Comfy CLI so AI agents and developers can generate images, video, audio, and 3D content without always opening the ComfyUI canvas.
Comfy Cloud MCP is in public beta. Setup, feedback channels, and related tools (Comfy Skills, Comfy CLI) are on the Comfy Cloud MCP page.

Comfy Cloud MCP

Connect to the hosted MCP server at cloud.comfy.org/mcp. Generate images, video, audio, and 3D on Cloud GPUs, search templates/models/nodes, run workflows, and share saved workflows from chat.

Comfy Partner MCP

Run a local MCP server using the Comfy Partner-Node SDK. Provides unified generate_image, generate_video, generate_3d, and more tools across 30+ partner providers.

Comfy CLI

Short agent-oriented overview of comfy generate. For the complete CLI guide, see Getting started.

Which One Should I Use?

Comfy Cloud MCPComfy Partner MCPComfy CLI
TypeRemote (hosted)Local MCP serverLocal CLI
RequiresComfy Cloud subscriptionComfy API keyComfy API key
ModelsComfy Cloud models (pre-installed)30+ partner providers (BFL, Ideogram, Kling, Runway, Veo, ElevenLabs, etc.)Partner nodes via comfy generate
WorkflowsFull ComfyUI workflow executionPartner API generation (no custom workflows)One-shot partner API calls (no workflow JSON)
GPUCloud GPUs (no local GPU needed)No GPU needed (API-based)No GPU needed (API-based)
Best forChat-driven agents, template searchUnified MCP tools across providersScripts, CI, batch jobs

What about local ComfyUI?

Many people ask about driving a local ComfyUI install with an agent instead of (or alongside) Comfy Cloud MCP. Why did Comfy Cloud MCP ship first? The hosted server at cloud.comfy.org is something only the Comfy team can build and operate. That is why our first-party MCP release is Cloud-based. Local agents are not new. Hooking an agent up to ComfyUI on your own machine has been possible since ComfyUI was open-sourced. A running local server exposes workflows and APIs that MCP bridges and other tools can call. Better local agent UX is on the roadmap. Improving the first-party local agent experience is part of our vision. Cloud MCP was the piece we could ship first; more local polish will follow. Community MCP servers you can try today. Several community projects already connect MCP clients to a local ComfyUI instance:
ProjectRepository
comfyui-mcpartokun/comfyui-mcp
comfyui-mcpshawnrushefsky/comfyui-mcp
comfyui-mcp-serverjoenorton/comfyui-mcp-server
comfy-mcp-serverlalanikarim/comfy-mcp-server
These are community-maintained. We link them for convenience; support, features, and compatibility vary by project. Want a terminal workflow instead of chat? See Comfy CLI (comfy generate, beta) for scripts and automation. To install and launch local ComfyUI from the terminal, start with Comfy CLI getting started.

What is MCP?

The Model Context Protocol (MCP) is an open standard that allows AI assistants to interact with external tools and services through a standardized interface. Instead of an AI agent needing to know custom API formats for every service, MCP provides a uniform way to expose tools. When you connect an MCP server to Claude Desktop, Claude Code, Cursor, or Amp, the AI assistant can:
  • Generate images from text descriptions
  • Create videos from text or images
  • Generate 3D models from text or images
  • Generate audio and music using AI models
  • Search models and templates to find the right tools for your task
…all through natural language conversation, without writing API calls manually.