Comfy Agent

Comfy Agent is a free agent skill maintained by Scopeful. It teaches an AI coding agent such as Claude Code, Cursor, Windsurf or Codex how to drive this tool correctly, so you do not have to re-explain it every session. Every published Scopeful skill is free and the install command is public, with no sign-in required. Install it with curl -H "Authorization: Bearer YOUR_API_KEY" -OJ https://scopeful.org/api/download?package=comfy-agent. Scopeful also tracks hand-verified USD pricing for 39 AI creative tools at https://www.scopeful.org/tools.

MCP server that gives AI coding agents direct control of ComfyUI, with a floating browser panel and shared self-learning memory across agents.

Tags: comfyui, mcp, agent, self-learning

Install

curl -H "Authorization: Bearer YOUR_API_KEY" -OJ https://scopeful.org/api/download?package=comfy-agent

Reference

name: comfy-agent description: MCP server that gives AI coding agents direct control of ComfyUI, with a floating browser panel and shared self-learning memory across agents.

What it does

Comfy Agent is an MCP server that connects AI coding agents (Claude Code, Codex, Cursor, Claude Desktop) to a running ComfyUI instance. Through 13 MCP tools, an agent can inspect workflows, create and connect nodes, set parameters, queue generations, view output images, and install custom nodes or models. Everything happens through the MCP protocol over stdio, no browser automation involved.

This is a different product from the comfyui-mcp toolkit. The comfyui-mcp toolkit is a headless Python script collection for batch workflow execution. Comfy Agent is an MCP server with a visual panel and persistent memory.

The window

Comfy Agent installs as a ComfyUI custom node. Once placed in custom_nodes/comfy-agent, it registers a floating dashboard panel inside ComfyUI's web interface (a round "CA" button in the bottom-right corner). The panel shows agent connection status, config snippets for each client, and live sync activity. No ComfyUI restart is needed for the MCP server itself, but the panel requires a reload to appear.

Self-learning memory

Every Comfy Agent instance reads and writes to a shared SQLite database at ~/.comfy-agent/memory.db, using WAL mode for lock-free concurrent reads. What gets saved:

When Claude Code discovers that FLUX guidance 3.5 works well for a particular style, it saves that via memory_save. Later, Cursor picks it up through memory_get_model_params without the user repeating anything. No daemon or sync service required. The file is the shared state.

Install

Download the zip, point a coding agent at the included INSTALL.md, and it handles everything: detecting your ComfyUI path, copying the custom node, registering the MCP server in your agent config, and verifying the connection. One prompt, fully automated.

The MCP server runs via uv run with no install step (dependencies are pulled on the fly). If uv is not available, a pip fallback (pip install mcp aiohttp) works the same way.

Requirements