Comfy MCP Integrates AI Agents with ComfyUI Workflows

ComfyUI released Comfy MCP, a Model Context Protocol server that bridges the gap between Large Language Models (LLMs) and node-based generative workflows. This integration allows AI agents to interact directly with the ComfyUI API to discover available nodes, construct graphs, and trigger image or video generations. By adopting the Model Context Protocol (MCP) standard, ComfyUI enables external tools to treat the generative engine as a functional extension of their own reasoning capabilities.

What's new

Comfy MCP functions as a standardized bridge that exposes the ComfyUI backend to any MCP-compliant client. As of November 2024, this includes support for Claude Desktop, Cursor, and various developer-centric AI environments. The server provides the agent with a list of installed nodes, their required inputs, and the ability to execute prompts.

Key technical capabilities include:

• Node Discovery: Agents can query the server to see which custom nodes are installed in a specific ComfyUI instance.
• Workflow Execution: LLMs can generate the JSON required to run a graph and receive the resulting media output.
• Context Awareness: The agent understands the specific parameters of a workflow, allowing it to troubleshoot errors or suggest optimizations based on the user's natural language request.

How it fits your workflow

Comfy MCP changes how technical artists and pipeline developers interact with complex node graphs. Instead of manually searching for the correct sampler or VAE loader, a user can instruct an agent like Claude to "create a workflow for a 2-second video using Stable Video Diffusion with a specific seed." This makes ComfyUI more accessible to those who understand the logic of generative AI but find the manual node-wiring process tedious.

For studio environments, Comfy MCP serves as a programmable interface for automation. It functions similarly to the API connectors found in tools like Leonardo.ai or RunComfy but offers deeper local control. While platforms like Krea or Magnific focus on simplified sliders, the Comfy MCP approach preserves the granular control of the node-based system while adding a natural language layer. This is particularly useful for rapid prototyping where an editor might need to generate dozens of variations of a background plate without leaving their coding environment or chat interface.

What it costs / how to try it

Comfy MCP is an open-source tool available via GitHub for users with an existing ComfyUI installation. It requires an MCP-compatible client, such as the Claude Desktop app or the Cursor code editor, to function as the interface.