1. Conduid
  2. Browser
  3. ConstantineB6/comfy-pilot
MCP server · Browser

ConstantineB6/comfy-pilot

MCP server + embedded terminal that lets Claude Code see and edit your ComfyUI workflows

77Good

Scored 3 hours ago · breakdown

About ConstantineB6/comfy-pilot

ConstantineB6/comfy-pilot is an MCP server published by ConstantineB6 in the Browser category: mCP server + embedded terminal that lets Claude Code see and edit your ComfyUI workflows. It has been installed 0 times through Conduid.

The repository has 119 stars and 25 forks, with the last commit 6 months ago. Six months or more without a commit doesn't mean the server is broken, but check the open issues (0) before depending on it in production.

Install

Install
npx comfy-pilot

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

Ask AI about ConstantineB6/comfy-pilot

Powered by Claude · Grounded in docs

I know everything about ConstantineB6/comfy-pilot. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ·README presentNot checked yet.
  • ·License declaredNot checked yet.
  • ·Tests presentNot checked yet.
  • ·Dependencies pinnedNot checked yet.
  • ·No dynamic code executionNot checked yet.
  • !Scoped permissionsDoesn't declare a permission scope. Assume it can do anything its process can.

Releases

v1.0.24v1.0.24 · 16 Feb 2026Current release

README

Comfy Pilot

Stars ComfyUI Registry

Talk to your ComfyUI workflows. Comfy Pilot gives Claude Code direct access to see, edit, and run your workflows — with an embedded terminal right inside ComfyUI.

Comfy Pilot

Why?

Building ComfyUI workflows means manually searching for nodes, dragging connections, and tweaking values one at a time. With Comfy Pilot, you just describe what you want:

  • "Build me an SDXL workflow with ControlNet" — Claude creates all the nodes, connects them, and sets the parameters
  • "Look at the output and increase the detail" — Claude sees your generated image and adjusts the workflow
  • "Download the FLUX schnell model and set up a workflow for it" — Claude downloads the model and builds a workflow from scratch

No copy-pasting node names. No hunting through menus. Just say what you want.

Installation

CLI (Recommended):

comfy node install comfy-pilot

ComfyUI Manager:

  1. Open ComfyUI
  2. Click ManagerInstall Custom Nodes
  3. Search for "Comfy Pilot"
  4. Click Install
  5. Restart ComfyUI

Git Clone:

cd ~/Documents/ComfyUI/custom_nodes && git clone https://github.com/ConstantineB6/comfy-pilot.git

Claude Code CLI will be installed automatically if not found.

Requirements

  • ComfyUI
  • Python 3.8+

Features

  • MCP Server - Gives Claude Code direct access to view, edit, and run your ComfyUI workflows
  • Embedded Terminal - Full xterm.js terminal running Claude Code right inside ComfyUI
  • Image Viewing - Claude can see outputs from Preview Image and Save Image nodes
  • Graph Editing - Create, delete, move, and connect nodes programmatically

Demo

https://github.com/user-attachments/assets/325b1194-2334-48a1-94c3-86effd1fef02

Usage

  1. Restart ComfyUI after installation
  2. The floating Claude Code terminal appears in the top-right corner
  3. The MCP server is automatically configured for Claude Code
  4. Ask Claude to help with your workflow:
    • "What nodes are in my current workflow?"
    • "Add a KSampler node connected to my checkpoint loader"
    • "Look at the preview image and tell me what you see"
    • "Run the workflow up to node 5"

MCP Tools

The MCP server provides these tools to Claude Code:

Tool Description
get_workflow Get the current workflow from the browser
summarize_workflow Human-readable workflow summary
get_node_types Search available node types with filtering
get_node_info Get detailed info about a specific node type
get_status Queue status, system stats, and execution history
run Run workflow (optionally up to a specific node) or interrupt
edit_graph Batch create, delete, move, connect, and configure nodes
view_image View images from Preview Image / Save Image nodes
search_custom_nodes Search ComfyUI Manager registry for custom nodes
install_custom_node Install a custom node from the registry
uninstall_custom_node Uninstall a custom node
update_custom_node Update a custom node to latest version
download_model Download models from Hugging Face, CivitAI, or direct URLs

Example: Creating Nodes

Create a KSampler and connect it to my checkpoint loader

Claude will use edit_graph to:

  1. Create the KSampler node
  2. Connect the MODEL output from CheckpointLoader to KSampler's model input
  3. Position it appropriately in the graph

Example: Viewing Images

Look at the preview image and describe what you see

Claude will use view_image to fetch and analyze the image output.

Example: Downloading Models

Download the FLUX.1 schnell model for me

Claude will use download_model to download from Hugging Face to your ComfyUI models folder. Supports:

  • Hugging Face (including gated models with token auth)
  • CivitAI
  • Direct download URLs

Terminal Controls

  • Drag title bar to move
  • Drag bottom-right corner to resize
  • Minimize
  • × Close
  • Reconnect session

Architecture

┌─────────────────────────────────────────────────────┐
│  Browser (ComfyUI)                                  │
│  ┌─────────────────┐  ┌──────────────────────────┐  │
│  │  xterm.js       │  │  Workflow State          │  │
│  │  Terminal       │  │  (synced to backend)     │  │
│  └────────┬────────┘  └────────────┬─────────────┘  │
│           │ WebSocket              │ REST API       │
└───────────┼────────────────────────┼────────────────┘
            │                        │
            ▼                        ▼
┌─────────────────────────────────────────────────────┐
│  ComfyUI Server                                     │
│  ┌─────────────────┐  ┌──────────────────────────┐  │
│  │  PTY Process    │  │  Plugin Endpoints        │  │
│  │  (claude CLI)   │  │  /claude-code/*          │  │
│  └─────────────────┘  └──────────────────────────┘  │
└─────────────────────────────────────────────────────┘
            │                        │
            │                        ▼
            │           ┌──────────────────────────┐
            └──────────▶│  MCP Server              │
                        │  (stdio transport)       │
                        └──────────────────────────┘

Files

  • __init__.py - Plugin backend: WebSocket terminal, REST endpoints
  • js/claude-code.js - Frontend: xterm.js terminal, workflow sync
  • mcp_server.py - MCP server for Claude Code integration
  • CLAUDE.md - Instructions for Claude when working with ComfyUI

Troubleshooting

"Command 'claude' not found"

Install Claude Code CLI:

macOS / Linux / WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell):

irm https://claude.ai/install.ps1 | iex

Windows (CMD):

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

MCP server not connecting

The plugin auto-configures MCP on startup. Check ComfyUI console for errors, or manually add to ~/.claude.json:

{
  "mcpServers": {
    "comfyui": {
      "command": "python3",
      "args": ["/path/to/comfy-pilot/mcp_server.py"]
    }
  }
}

Terminal disconnected

Click the ↻ button to reconnect, or check ComfyUI console for errors.

License

MIT

README mirrored from the source repository 3 hours ago. The original is authoritative.

Questions

About ConstantineB6/comfy-pilot

How do I install ConstantineB6/comfy-pilot?

Run npx comfy-pilot, then add the server to your MCP client's configuration. Conduid has recorded 0 installs, so the command is known to work with current clients.

Is ConstantineB6/comfy-pilot safe to use with an AI agent?

Its trust score is 77 out of 100 (good). It passes 0 of 1 static security checks; the failures are listed above. It has no ConduID identity yet, so agent calls to it are not receipted.

Is ConstantineB6/comfy-pilot still maintained?

The last commit was 6 months ago, with 0 open issues. That's long enough that you should check whether the maintainer is responding to issues before depending on it.