comfyui-mcp-panel

Name: comfyui-mcp-panel
Author: artokun

Verified

The local-first sidebar AI agent for ComfyUI — runs on your own Claude OR ChatGPT subscription (no API keys, no extra LLM costs). Drives your live graph: edits, workflows & installs in natural language. The panel UI for comfyui-mcp, the agent-native control plane for ComfyUI.

101stars

21forks

JavaScript

Installation

# Add to your Claude Code skills
git clone https://github.com/artokun/comfyui-mcp-panel

Getting Started

Guides for using ai agents skills like comfyui-mcp-panel.

Security ReportVerified

Last scanned: 8/27/2026

{
  "issues": [],
  "status": "PASSED",
  "scannedAt": "2026-08-27T15:02:31.429Z",
  "npmAuditRan": true,
  "pipAuditRan": true,
  "promptInjectionRan": true
}

README.md

Frequently Asked Questions

What is comfyui-mcp-panel?

comfyui-mcp-panel is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by artokun. The local-first sidebar AI agent for ComfyUI — runs on your own Claude OR ChatGPT subscription (no API keys, no extra LLM costs). Drives your live graph: edits, workflows & installs in natural language. The panel UI for comfyui-mcp, the agent-native control plane for ComfyUI. It has 101 GitHub stars.

Is comfyui-mcp-panel safe to use?

Yes. comfyui-mcp-panel passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.

How do I install comfyui-mcp-panel?

Clone the repository with "git clone https://github.com/artokun/comfyui-mcp-panel" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is comfyui-mcp-panel written in?

comfyui-mcp-panel is primarily written in JavaScript. It is open-source under artokun on GitHub, so you can review or fork the full source.

Are there alternatives to comfyui-mcp-panel?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh comfyui-mcp-panel against similar tools.

Agentic AI for Beginners

Build your first AI agent from scratch - tool use, ReAct pattern, memory, deployment

41 minBeginner

Comments (0)

to leave a comment.

No comments yet. Be the first to share your thoughts!

Related Skills

ECC

by affaan-m

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

243,639

claude-emporium daofy

⚠️ Third-Party Software Notice

This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.

Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.

Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.

Read the Terms of Service

ComfyUI Agent Panel

📦 On ComfyUI-Manager & the Comfy Registry as comfyui-agent-panel

The polished public release — native ComfyUI design system, live activity cards for every agent edit, and multi-tab support. Search comfyui-agent-panel in ComfyUI-Manager to install, or see the docs.

An autonomous AI agent embedded in the ComfyUI sidebar that drives your canvas — now on EITHER Claude or ChatGPT (your own subscription, no API key).

One-click GPU pod — a ready-to-run ComfyUI with this panel + comfyui-mcp + ComfyUI-Manager v2 preinstalled, on your own GPU. No setup.

Stuck or have a question? Join the Discord — or hit 🆘 Need help? in the panel's Settings → About (it copies a diagnostics summary and opens the Discord for you).

🌐 Speaks your language

The whole panel — chat, settings, the CivitAI browser, the training wizard, the RunPod controls, every tooltip — is translated into 12 languages, and it follows ComfyUI's own language setting automatically. Change it any time in Settings → Panel language.

English · 한국어 · 日本語 · 中文 (简体) · 中文 (繁體) · Русский · Français · Español · Português (BR) · Türkçe · العربية · فارسی

Arabic and Persian render right-to-left. Counted strings use each language's real plural rules — Russian's four forms and Arabic's six, not an English one/other pair bolted on.

Pick a provider — Claude or ChatGPT — and the matching agent runs in the background on your subscription, sees the graph you're looking at, and edits it live. Every provider is offered the same surface: the same live-canvas tools, the same model knowledge, the same one-shot workflow loads, the same cost guardrail. Offered, not guaranteed received — a provider applies its own tool budget to what we hand it, and Codex was seen dropping the live-canvas tools silently (#291). That is fixed upstream, and the panel now asks the agent to say so rather than improvise if it ever recurs — see What the agent can do.

Part of the comfyui-mcp project — the local-first, agent-native control plane for ComfyUI (MCP server + agent orchestrator). Full documentation at comfyui-mcp.artokun.io/docs.

Type "add a KSampler and wire it to my checkpoint" in the panel and watch nodes appear on the canvas. Every edit is undoable with Ctrl+Z.

No API keys. No extra LLM costs. The agent runs on your Claude or ChatGPT subscription — a background comfyui-mcp orchestrator the panel starts for you when you click Connect. The panel is just its window into your graph.

this panel ⇄ loopback bridge ⇄ comfyui-mcp orchestrator (background, Claude OR ChatGPT — your subscription) ⇄ your graph

Every provider shares one orchestrator on one loopback port (default ws://127.0.0.1:9199; 9180 is a legacy fallback), so you pick a provider rather than juggle ports.

Features

Capability	What it does
Pick a provider	A backend picker with Claude / ChatGPT chips — choose the agent, not a port. Switching providers starts a fresh chat (sessions aren't shared across providers) and posts a system note.
Provider onboarding	Connect-time readiness detection per provider (CLI on PATH + a login on disk; macOS Keychain handled). An onboarding card shows only when neither provider is signed in; the panel auto-switches to a ready provider when your saved pick isn't usable (your saved preference is untouched), and a not-ready row becomes a "set up" action.
Live-canvas building	The agent adds, wires, moves, retitles, colors, collapses, groups, and lays out nodes on the graph you're viewing — all through a fixed `panel_` allowlist (no arbitrary JS), every edit undoable with Ctrl+Z*.
One-shot workflow / pack load	`panel_load_workflow` drops a whole graph onto the canvas in one call — load a bundled installer pack's local-GPU workflow by name without shuttling the JSON through the chat.
Local-GPU vs paid-API awareness	Bundled packs are local/free; for ad-hoc graphs the agent checks the runtime (`list_packs` `action:"check_runtime"`) and asks before spending paid API credits.
Installer packs + skills	The agent discovers bundled model-family skills and one-command installer packs, then applies the manifest and loads the ready workflow instead of hand-building a graph.
Rewind & rollback	Roll back code (graph), conversation (fork the session), or both from any past message, plus `/revert` and double-Esc quick rewind.
Autonomous install → restart → continue	Install custom nodes through your own ComfyUI Manager, restart ComfyUI to load them, and the panel auto-reconnects so the agent resumes its task.
Reasoning effort + model selector	A per-provider effort/model picker — Claude's models, or ChatGPT's GPT-5-class models via Codex — moved into the model selector; a chosen effort survives a provider switch by mapping to the nearest valid level.
Pending-message tray + reconnect durability	Queue messages while the agent is busy (edit / send-now / reorder), and reclaim a wedged orchestrator on Connect.
Apps (micro-apps)	Convert any workflow into a named, one-click app (name, description, thumbnail, chosen input/output endpoints) from the toolbar's Apps button — runs headless (the canvas is never touched), locally or on a connected RunPod pod, with best-effort hidden-workflow packaging and a publish/explore registry (`registry/`, Cloudflare Worker + D1 + R2) shared with the mobile app.
Durable workflow chat history	Keep multiple named or pinned chats per workflow, search across them, export/import JSON, survive ComfyUI/browser restarts through IndexedDB, and track the graph version used by every user turn.

Quickstart

Install the pack — search comfyui-agent-panel in ComfyUI-Manager, or git clone https://github.com/artokun/comfyui-mcp-panel into ComfyUI/custom_nodes. Restart ComfyUI; an Agent tab (💬) appears in the sidebar.
Sign in to your provider once so the agent can run on your subscription (Node ≥ 22):
- Claude (via the Claude CLI): claude (or claude setup-token)
- ChatGPT (via the Codex CLI): codex login
Open the Agent tab, pick Claude or ChatGPT in the backend picker, and click Connect. The panel starts that provider's background orchestrator on your subscription — no API keys — and the status pill turns green.
Type a request — "build a Flux txt2img graph and run it" — and watch the edits land on your canvas. Disconnect stops the agent; nothing is ever started without your click.

Install

Via ComfyUI-Manager (recommended): search for comfyui-agent-panel and install.

Via git:

cd ComfyUI/custom_nodes
git clone https://github.com/artokun/comfyui-mcp-panel

Restart ComfyUI. A new Agent tab (💬) appears in the sidebar.

Connect

claude        # Claude — or: claude setup-token
codex login   # ChatGPT (Codex)

Open the Agent tab, pick a provider in the backend picker (Claude / ChatGPT chips), and click Connect. The panel starts that provider's autonomous background agent — npx -y comfyui-mcp@latest connect, running on your subscription (no API keys) — and the status pill turns green. Type a request ("build a Flux txt2img graph and run it") and watch the edits land on your canvas. Disconnect stops the agent; nothing is ever started without your click.

Switching providers starts a fresh chat — conversations aren't shared across Claude and ChatGPT — and the panel tells you so. Every provider shares one orchestrator on one loopback port (default ws://127.0.0.1:9199, overridable via COMFYUI_MCP_BRIDGE_PORT; 9180 is a legacy fallback so a live session there is not stranded). The bridge is loopback-only. To run an orchestrator yourself, set COMFYUI_MCP_NO_AUTOSPAWN=1, launch it manually, then click Connect (the Bridge URL lives under Advanced).

Type / in the composer for commands — panel ones like /reload (pick up new code, keep the chat), /reload-ui (reload just the panel), /revert (undo the last turn's graph edits), /record-skill (save the open graph as a reusable skill), and /restart (recover an unresponsive agent — kills the orchestrator and its child tree, starts fresh). On the Claude backend, provider slash commands (/compact, /loop, …) are available too.

What the agent can do

The agent drives the workflow you're viewing through a fixed allowlist of panel_* commands (no arbitrary JavaScript). Every graph mutation goes through LiteGraph's change tracking, so ComfyUI's native Ctrl+Z reverts an agent edit exactly like your own. The panel_* tools live in one shared definition list, registered onto the in-process Claude Agent SDK server and a loopback HTTP MCP the orchestrator hosts for the ChatGPT/Codex and other CLI backends, so both surfaces are identical as served.

They are not automatically identical as received, and this page used to claim they were. A backend applies its own tool budget to everything