# ComfyUI MCP > Autonomous AI agent + MCP server for ComfyUI — drive your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama, or any hosted model (DeepSeek, GLM, MiMo, OpenRouter). Generate images and video, author and run workflows, manage models and custom nodes. Includes an LLM Arena that benchmarks models on real ComfyUI tasks, and a Claude Code plugin with skills, slash commands, and installer packs. ## English ### Guides #### Getting Started - [ComfyUI MCP](https://comfyui-mcp.artokun.io/docs/index.md): The local-first, agent-native control plane for ComfyUI — an MCP server + Claude Code plugin that generates images, video and audio, authors and runs workflows, and manages models and custom nodes in natural language. Local, remote, or Comfy Cloud. - [Quickstart](https://comfyui-mcp.artokun.io/docs/quickstart.md): Add ComfyUI MCP to Claude Code and generate your first image. - [Using the tools](https://comfyui-mcp.artokun.io/docs/using-tools.md): What the tools are, why you never call one yourself, and what to do when one says no. Written for people, not for engineers. - [Local vs. Comfy Cloud agent](https://comfyui-mcp.artokun.io/docs/local-vs-comfy-cloud.md): Comfy-Org's In-App Agent and Comfy Cloud MCP run on Comfy Cloud; comfyui-mcp runs on your own ComfyUI install, on any LLM — Claude or ChatGPT on your existing subscription, or a free local model via Ollama with no account and no network. An honest comparison of when to use which, including the cases… - [Installation](https://comfyui-mcp.artokun.io/docs/installation.md): Install ComfyUI and connect the MCP server — local, remote, or Comfy Cloud. - [Roadmap](https://comfyui-mcp.artokun.io/docs/roadmap.md): What's shipped, what's being built, what's next — and what we've decided not to build, with the reasoning. Ordered by priority, not by date. #### Configuration - [Configuration](https://comfyui-mcp.artokun.io/docs/configuration.md): Environment variables, transports, and connection targeting. - [Remote / hosted connector](https://comfyui-mcp.artokun.io/docs/remote-connector.md): Expose comfyui-mcp as an authenticated, publicly-reachable Streamable-HTTP MCP server — add it to Claude Desktop's Custom Connectors or any remote client, one command. - [Cloud deployment (RunPod)](https://comfyui-mcp.artokun.io/docs/cloud-deployment.md): Deploy, connect, monitor, and stop a cloud GPU pod for ComfyUI without leaving the Agent Panel — one-tap deploy, an honest local⇄pod host switch, live cost/GPU status, and idle auto-stop. Or drive an existing pod with one command. - [Docker & Compose](https://comfyui-mcp.artokun.io/docs/docker.md): Run ComfyUI as a docker-compose service and connect comfyui-mcp to it — and why comfyui-mcp itself is never a compose service. - [Self-hosted secure bridge relay (enterprise)](https://comfyui-mcp.artokun.io/docs/self-hosted-relay.md): Own the full loop: run your own relay for the secure wss:// bridge instead of depending on a third-party service — for data residency, compliance, or a no-external-dependency requirement. - [How it works](https://comfyui-mcp.artokun.io/docs/concepts.md): The model behind the tools — transports, the ComfyUI-Manager API, and the local/remote/cloud modes. - [Claude Code Plugin](https://comfyui-mcp.artokun.io/docs/plugin.md): comfyui-mcp is a full Claude Code plugin: 42 AI skills, 11 slash commands, 4 autonomous agents, and 3 hooks on top of the 38 MCP tools — expert ComfyUI knowledge that grows with every release. - [Sidebar Panel](https://comfyui-mcp.artokun.io/docs/panel.md): ComfyUI Agent Panel — an autonomous AI agent in ComfyUI's sidebar that drives your canvas on ANY LLM: Claude, ChatGPT, or Gemini on your subscription (no API key), a free local model via Ollama (no account), or any hosted model over an OpenAI-compatible endpoint. Available on the Comfy Registry as c… - [Apps (micro-apps)](https://comfyui-mcp.artokun.io/docs/apps.md): Turn a workflow into a one-click app: a manifest, an exposed run form, and an API prompt snapshot that values are patched into per run. Convert in the panel, run from the panel, the phone, or an agent, and publish to a public registry. - [Mobile app (beta)](https://comfyui-mcp.artokun.io/docs/mobile.md): ComfyUI-MCP Mobile — chat with your ComfyUI agent from your phone. Pair with the Agent Panel's QR code and drive generations, workflows, and models from anywhere on your network. Now in closed beta on Android and iOS. - [Third-party hosts](https://comfyui-mcp.artokun.io/docs/third-party-hosts.md): Build your own front-end (Blender panel, browser extension, another app) on the same pairing protocol the mobile app uses — one agent, shared context, thin client. The message shapes, endpoints, security invariants, and a copy-paste prompt to have an LLM scaffold your adapter. - [Backends / providers](https://comfyui-mcp.artokun.io/docs/backends.md): The panel agent runs on ANY LLM: Claude, ChatGPT, Gemini, Grok, Kimi, or GLM on your own subscription/plan, a free local model via Ollama / LM Studio / llama.cpp (no account at all), or any hosted model over an OpenAI-compatible endpoint. How the provider-neutral AgentBackend port, the picker, and t… - [Local LLMs & other agents](https://comfyui-mcp.artokun.io/docs/local-llms.md): First-class support for Hermes Agent, OpenClaw, and Copilot CLI — plus any MCP client on a local (Ollama) or hosted model. One-command setup, and a compact tool mode that collapses ~200 tool schemas into 3 meta-tools. - [LLM Arena](https://comfyui-mcp.artokun.io/docs/arena.md): Benchmark ANY agent LLM on real ComfyUI tasks — local (Ollama), hosted (OpenRouter/DeepSeek/GLM/MiMo), or frontier. Every score is verified against the ComfyUI server, not the model's claims. One command, share-ready reports. - [Troubleshooting](https://comfyui-mcp.artokun.io/docs/troubleshooting.md): Fixes for the issues users actually hit: ComfyUI-Manager version mismatches (405s), git-URL installs being silently skipped, the panel unreachable from a remote browser, stale npx caches, and port-forwarded remotes misdetected as local. #### Tools - [kitchen and panel_kitchen](https://comfyui-mcp.artokun.io/docs/kitchen.md): See what comfy-kitchen can do on this GPU, find where the open graph is leaving it on the table, and apply the faster path with proof. ### Tool Reference #### Tools - [Image & Audio Generation](https://comfyui-mcp.artokun.io/docs/tools/image-generation.md): High-level text-to-image, audio, video and 3D generation, conditioned image variants, and the two post-processing passes — all nine actions of one tool since 0.50.0 slice 16. - [Workflow Execution](https://comfyui-mcp.artokun.io/docs/tools/workflow-execution.md): Enqueue workflows — one at a time, from a named template, or as a batch — and inspect the queue, jobs, history, and system stats. - [Workflow Authoring](https://comfyui-mcp.artokun.io/docs/tools/workflow-authoring.md): Build, modify, validate, and visualize ComfyUI workflows. - [Workflow Library](https://comfyui-mcp.artokun.io/docs/tools/workflow-library.md): Save, load, strip/slice, analyze, and extract workflows. - [Assets & Images](https://comfyui-mcp.artokun.io/docs/tools/assets-images.md): View, convert, and upload generated images; analyze colors; stage outputs as inputs; upload media inputs; browse outputs. - [Models](https://comfyui-mcp.artokun.io/docs/tools/models.md): Search (HuggingFace + CivitAI), download, list, and remove models; resolve a workflow's missing models with VRAM-aware candidates; manage embeddings and VRAM. - [Custom Nodes](https://comfyui-mcp.artokun.io/docs/tools/custom-nodes.md): Discover, install, update, snapshot, bisect, scaffold, and publish custom node packs. - [API Nodes](https://comfyui-mcp.artokun.io/docs/tools/api-nodes.md): Discover and run hosted partner / API nodes (comfy.org). - [Install & Environment](https://comfyui-mcp.artokun.io/docs/tools/install-environment.md): Install/update ComfyUI and the sidebar panel, self-update the MCP server, apply a setup manifest, manage workspaces, inspect the environment, configure ComfyUI-Manager, report issues. - [Process Control](https://comfyui-mcp.artokun.io/docs/tools/process-control.md): Start, stop, and restart the ComfyUI process. - [Defaults](https://comfyui-mcp.artokun.io/docs/tools/defaults-stats-skills.md): Generation defaults and ComfyUI frontend UI settings. Stats now live on get_system_stats and get_history; skills on the Skills & Knowledge page. - [RunPod (cloud GPU)](https://comfyui-mcp.artokun.io/docs/tools/runpod.md): Deploy, connect, monitor, and stop RunPod cloud GPU pods, and switch rendering between your local rig and a pod — the tools behind the panel/mobile RunPod control panel. - [LoRA Training](https://comfyui-mcp.artokun.io/docs/tools/training.md): Train character LoRAs (FLUX.1-dev) via ostris ai-toolkit — locally in a GPU Docker image or on a rented RunPod pod — with a crash-safe job registry and streamed progress. - [Apps (micro-apps)](https://comfyui-mcp.artokun.io/docs/tools/apps.md): List, inspect, and run the panel's micro-apps — named, one-click workflow apps with exposed inputs — for canvas-less clients (mobile). - [comfy-cli](https://comfyui-mcp.artokun.io/docs/tools/comfy-cli.md): Drive the official comfy-cli — managed server lifecycle, jobs, loaded-node search, workflow validation/execution, uploads/downloads, model discovery, and official agent skills. - [Skills, Packs & Runtime Cost](https://comfyui-mcp.artokun.io/docs/tools/skills-knowledge.md): The list_packs knowledge tool — bundled ComfyUI expertise (skills, installer packs, workflow templates) for any agent backend, the local-GPU-vs-paid-API cost guardrail, and the custom-node dependency check/install. ### Blog #### Blog - [Blog](https://comfyui-mcp.artokun.io/docs/blog/index.md): Notes on building comfyui-mcp — tool design, quality, and a model-highlight series on the open image & video models you can run locally. - [comfy-kitchen in ComfyUI: see the faster path, then prove it](https://comfyui-mcp.artokun.io/docs/blog/comfy-kitchen.md): kitchen and panel_kitchen report what comfy-kitchen can do on this GPU, name the UNETLoader still on default dtype, and apply fp8 / INT8 attention / NVFP4 with a before/after run — or revert. - [From bug report to published release, autonomously](https://comfyui-mcp.artokun.io/docs/blog/from-report-to-release.md): How comfyui-mcp turns a bug report into a shipped fix: an MCP intake tool, a Cloudflare Worker that triages with an AI agent, a swarm of coding agents, an adversarial merge gate, and a tagged release. - [Micro-Apps: turn a ComfyUI workflow into a form anyone can run](https://comfyui-mcp.artokun.io/docs/blog/micro-apps.md): A workflow is a graph, and a graph is a hostile interface for someone who just wants to change three things and hit Go. Apps package a workflow as a manifest plus an API-format prompt snapshot, generate a run form from it, and expose the whole thing over apps_* MCP tools — so the panel, the phone, a… - [ComfyUI on RunPod: Rent a Cloud GPU](https://comfyui-mcp.artokun.io/docs/blog/runpod-comfyui.md): Run ComfyUI on a RunPod cloud GPU with one-tap deploy, custom nodes and models, clear host switching, live pricing, and idle auto-stop cost control. - [Train a FLUX LoRA in the Cloud on RunPod](https://comfyui-mcp.artokun.io/docs/blog/train-lora-runpod.md): Train a FLUX.1-dev character LoRA on a RunPod cloud GPU — no 24GB card required. Upload your dataset by SSH and get a ComfyUI-ready .safetensors file. - [Train a character LoRA by asking: the train_* tools land](https://comfyui-mcp.artokun.io/docs/blog/lora-trainer-p1.md): comfyui-mcp's panel agent can now drive LoRA training end to end — dataset staging, Docker + GPU preflight, a headless ai-toolkit container, streamed progress, and honest cancel — all from a conversation. P1 ships character LoRAs on FLUX.1-dev, E2E-proven on an RTX 4090. - [Blind mode said the agent NEVER sees your images. A user proved it was lying — here's the mechanical fix](https://comfyui-mcp.artokun.io/docs/blog/blind-mode-privacy.md): The panel's Blind toggle promised the AI agent never receives image pixels — but get_image and view_image fetched them straight from ComfyUI's /view anyway. How issue #90 exposed a promptable privacy promise, and how one wrapper at the tool-registration boundary made it a mechanical guarantee in com… - [GPT-5.6 lands in the ComfyUI agent — Sol, Terra, and Luna on your ChatGPT subscription](https://comfyui-mcp.artokun.io/docs/blog/gpt-5-6-comfyui.md): comfyui-mcp v0.41.0 brings OpenAI's GPT-5.6 family — Sol, Terra, and Luna — to the ComfyUI Agent Panel through the Codex integration, on your own ChatGPT subscription with no API key. The new max and ultra reasoning efforts, a live account-aware model catalog that only shows what YOUR plan can run,… - [Kimi K3: the biggest open-weight model ever now drives your ComfyUI canvas](https://comfyui-mcp.artokun.io/docs/blog/kimi-k3-comfyui.md): Moonshot AI's Kimi K3 — 2.8 trillion parameters, a 1M-token context window, and open weights promised for July 27 — is the closest an open model has ever come to the closed frontier. Why the release matters, what the benchmarks and the caveats actually say, and how to run it as your ComfyUI agent to… - [The 405 that wasn't: one Manager API, three dialects, two excellent bug reports](https://comfyui-mcp.artokun.io/docs/blog/manager-405-dialects.md): comfyui-mcp installs custom nodes through ComfyUI-Manager's HTTP API — which turned out to exist in three incompatible dialects. A debugging war story in three acts: the released 3.x Manager vs the pip v4 package, a Docker flag that swaps the server under your feet, and the aiohttp catchall that mak… - [Panel-owned sessions: your conversation now survives workflow switching](https://comfyui-mcp.artokun.io/docs/blog/panel-owned-sessions.md): The comfyui-mcp Panel agent used to treat every ComfyUI workflow tab as a separate conversation — switch, save, or rename a workflow and your chat emptied mid-task. Now the conversation is the unit of continuity: the orchestrator rebinds the live agent across tabs (no respawn, so even in-memory Olla… - [CivitAI, Inside ComfyUI: Browse, Like, Load, Recreate](https://comfyui-mcp.artokun.io/docs/blog/civitai-in-comfyui.md): The panel now ships a full CivitAI browser in ComfyUI's sidebar — images, videos, checkpoints, LoRAs, and workflows behind a server-side proxy with PKCE OAuth, a generation-details lightbox, first-class likes, creator search qualifiers, and 'load this workflow onto my canvas' as a single click. - [The ComfyUI mobile app: your agent, your renders, your CivitAI — from your phone](https://comfyui-mcp.artokun.io/docs/blog/comfyui-mobile-app.md): The ComfyUI-MCP mobile companion (Flutter, iOS + Android beta) puts the same autonomous agent that drives your desktop ComfyUI in your pocket: chat with it, watch renders land, browse CivitAI with full likes/collections/creator parity, pick providers, and even mirror and drive a live desktop tab. Pa… - [Untangling Workflow Spaghetti: Flatten Get/Set Buses and Use-Everywhere Links Without Wrecking the Layout](https://comfyui-mcp.artokun.io/docs/blog/flatten-workflows.md): Community ComfyUI workflows arrive as 128-node graphs wired through rgthree Get/Set buses, Reroutes, and cg-use-everywhere broadcasts. Three new tools — panel_strip_workflow, panel_slice_workflow, and panel_flatten_workflow — resolve the virtual wiring into real links so agents (and humans) can actu… - [The orchestrator that never goes stale: self-updating, self-restarting, and the art of waiting for idle](https://comfyui-mcp.artokun.io/docs/blog/self-updating-orchestrator.md): A long-running panel orchestrator quietly falls behind npm releases — users file bugs that shipped fixed weeks ago, and developers debug 'regressions' that are really a days-old process serving a fresh checkout. In v0.34.0 the orchestrator updates itself: hourly npm re-checks, restart-on-rebuild for… - [We fine-tuned Gemma 4 to drive ComfyUI — then built an Arena that told us our smallest model was broken](https://comfyui-mcp.artokun.io/docs/blog/gemma4-fine-tune-arena.md): The gemma4-comfyui-mcp ladder (e2b/e4b/12b) is Gemma 4 fine-tuned on server-verified ComfyUI tool trajectories — and the LLM Arena that scores it is honest enough to have caught our own regression. How the v1 e2b scored 4/20 by emitting call_tool without a tool name, why the fix was a training-data… - [Giving every backend eyes — and making them admit when they're blind](https://comfyui-mcp.artokun.io/docs/blog/vision-per-model.md): An image-generation agent that can't SEE what it generated is flying blind. How the ComfyUI agent delivers vision on every backend — Claude, ChatGPT, Gemini, Ollama, OpenRouter, and friends — by treating vision as a property of the MODEL, not the provider: always attempt delivery, and on rejection d… - [Bring any LLM: free local models now drive ComfyUI](https://comfyui-mcp.artokun.io/docs/blog/local-llms-comfyui.md): The ComfyUI agent panel now runs on Ollama, LM Studio, llama.cpp, or any OpenAI-compatible endpoint — no account, no API key, fully offline. How a 6-tool compact router makes the full tool catalog survivable for a 4B model, the reliability work small models actually needed (loop-breakers, empty-fina… - [The ComfyUI agent now runs on Claude OR ChatGPT — pick a provider, not a port](https://comfyui-mcp.artokun.io/docs/blog/comfyui-agent-claude-or-chatgpt.md): The autonomous AI agent in ComfyUI's sidebar that drives your canvas now runs on EITHER Claude or ChatGPT — on your own subscription, no API keys. How an AgentBackend port (dependency injection) gives both providers full parity: live-canvas panel_* tools served as a loopback MCP, Codex driven over c… - [The self-healing ComfyUI agent: it crashed at 99%, then fixed the node that crashed it](https://comfyui-mcp.artokun.io/docs/blog/self-healing-comfyui-agent.md): A Wan2.2 render hard-crashed ComfyUI with a native access violation (0xC0000005) at 99% — a fp8 LoRA merge inside ComfyUI-WanVideoWrapper. The autonomous ComfyUI agent read the faulthandler dump on restart, named the culprit node and file:line, updated it, retried, patched it, and offered the fix up… - [The panel that talks back: an autonomous agent on your Claude subscription](https://comfyui-mcp.artokun.io/docs/blog/panel-agent-subscription.md): Getting the ComfyUI sidebar panel driven by a real background agent — one that runs on your Claude subscription with no API key, takes messages over a live channel, and can be interrupted mid-task — took three tries. Push notifications can't wake an idle session. The --sdk-url transport that did eve… - [Giving the ComfyUI Panel Agent Eyes — and a Sense of Layout](https://comfyui-mcp.artokun.io/docs/blog/panel-workflow-layout.md): How we taught the comfyui-mcp Panel agent to organize ComfyUI workflows: pos/size-aware reads, subgraph rail control, group boxes, node color/collapse, and a screenshot tool so it can actually see the canvas it's building. - [Installer packs that can't rot: one manifest, every install](https://comfyui-mcp.artokun.io/docs/blog/installer-packs-that-cant-rot.md): The community '1-click installer' .bat is great UX and a maintenance trap — Windows-only, duplicated in a separate .sh, and full of model URLs that rot silently. Here's how comfyui-mcp turns an install into a single manifest that drives both the double-click scripts and an MCP-native install, with C… - [From B to A: sharpening 70+ tool descriptions with TDQS](https://comfyui-mcp.artokun.io/docs/blog/comfyui-mcp-tdqs-case-study.md): A case study using Glama's Tool Definition Quality Score to improve comfyui-mcp's 70+ tools. #### Model Highlights - [How to Run Krea 2 Locally in ComfyUI](https://comfyui-mcp.artokun.io/docs/blog/krea2-comfyui.md): Run Krea 2 Turbo — Krea.ai's 12B open-weight text-to-image model — locally in ComfyUI. 8-step turbo speed, Qwen3-VL prompting, JSON area prompts, install, and VRAM tips. - [WAN 2.2 in ComfyUI: The Open Video King](https://comfyui-mcp.artokun.io/docs/blog/wan-2.2-comfyui.md): Run WAN 2.2 — the open-source MoE video model — locally in ComfyUI. High/low-noise experts, I2V & T2V, GGUF VRAM tiers, Lightning, one-command install. - [Qwen-Image & Qwen-Image-Edit in ComfyUI](https://comfyui-mcp.artokun.io/docs/blog/qwen-image-comfyui.md): Run Qwen-Image 20B and Qwen-Image-Edit locally in ComfyUI: instruction editing, manual-mask inpainting, the WAN 2.2 refine combo, and GGUF VRAM tiers. - [WAN Animate 2.2 in ComfyUI: Character Replace](https://comfyui-mcp.artokun.io/docs/blog/wan-animate-comfyui.md): Replace any character in a video from ONE reference image, or transfer motion to a still, with WAN Animate 2.2 in ComfyUI. Install, VRAM, fixes. - [WAN Transparent Expressions in ComfyUI: Alpha Sprites](https://comfyui-mcp.artokun.io/docs/blog/wan-transparent-comfyui.md): Make looping, transparent-background animated character expressions with WAN 2.2 I2V + BiRefNet in ComfyUI — RGBA WebP/GIF sprites for SillyTavern. - [Z-Image in ComfyUI: The Low-VRAM Speed King](https://comfyui-mcp.artokun.io/docs/blog/z-image-comfyui.md): Run Z-Image Turbo and Base — Alibaba's 6B text-to-image model — locally in ComfyUI under 8 GB VRAM. Install, GGUF tips, LoRA training, and the V2 ULTRA workflow. - [Run ERNIE-Image Locally in ComfyUI (Apache-2.0)](https://comfyui-mcp.artokun.io/docs/blog/ernie-image-comfyui.md): Run Baidu's ERNIE-Image — the Apache-2.0, 8B text-to-image model with sharp multilingual EN/CN/JP text — locally in ComfyUI. Install, settings, VRAM. - [ANIMA 1.0: The Tiny Anime Model for ComfyUI (Under 6 GB)](https://comfyui-mcp.artokun.io/docs/blog/anima-comfyui.md): ANIMA 1.0 is a ~2B anime text-to-image model for ComfyUI. Danbooru tags + natural language, anime inpainting, and LoRA training — all under 6 GB VRAM. - [LTX-2.3 in ComfyUI: Fast GGUF Video + LTX Director](https://comfyui-mcp.artokun.io/docs/blog/ltx-2.3-comfyui.md): Run LTX-2.3 — Lightricks' 22B GGUF text-to-video & image-to-video model — locally in ComfyUI. Install, VRAM tiers, LTX Director timeline, and the kornia fix. - [ComfyUI Video Upscaling: SeedVR2, FlashVSR & RIFE](https://comfyui-mcp.artokun.io/docs/blog/video-upscale-comfyui.md): Upscale and restore AI video locally in ComfyUI — SeedVR2 and FlashVSR temporal restorers, the downscale-first trick, the quick per-frame ESRGAN path, and built-in RIFE frame interpolation. Pipeline order, VRAM tiers, and one-command setup. - [Video Extension in ComfyUI: Pusa 2.2 Temporal Flowmatching](https://comfyui-mcp.artokun.io/docs/blog/video-extend-pusa-comfyui.md): Extend AI video in ComfyUI with Pusa 2.2 — temporal flowmatching that continues a WAN 2.2 clip instead of regenerating it. Pipeline, models, LoRAs, settings, and the fp8 gotcha. - [How to Run Ideogram 4 Locally in ComfyUI](https://comfyui-mcp.artokun.io/docs/blog/ideogram-4-comfyui.md): Run Ideogram 4 — the 9.3B open-weight text-to-image model — locally in ComfyUI. Install, VRAM tips, JSON area prompting, and flawless readable text. ### Changelog #### Changelog - [Changelog](https://comfyui-mcp.artokun.io/docs/changelog/index.md): What's new in comfyui-mcp. - [July 28: a security patch, reconnect self-heal, and OpenRouter's full catalog](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-28.md): Path-traversal hardening on the upload tools, a panel session that self-heals after a full reconnect/reload/workflow-switch instead of getting stuck on a dead tab, the complete OpenRouter model list, a Windows restart fix, and a third-party host protocol guide. - [July 21: rent a GPU from the panel — RunPod connector and cloud LoRA training](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-21.md): A first-party RunPod connector deploys and drives a cloud GPU pod from the panel and phone with an honest local⇄pod host switch, live cost/GPU status, and idle auto-stop; the LoRA trainer learns to train on rented pods; plus resolve_missing_models and provider model discoverability. - [July 20: GPT-5.6, the LoRA trainer lands, Blind mode gets mechanical](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-20.md): The GPT-5.6 family (Sol / Terra / Luna) arrives with the extended max/ultra effort scale; seven train_* tools train character LoRAs on FLUX.1-dev end-to-end from chat; the panel's Blind toggle now mechanically gates every image-returning tool; and the Gemini backend is revived with a self-healing mo… - [July 19: see what I see — viewport tools and one honest error surface](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-19.md): New panel tools let the agent read your current selection and viewport, and panel_get_errors becomes the single error surface — every errored node joined to its cause (missing model with download URL, missing media, validation, runtime exception), closing the 'red node, no error message' gap. - [July 18: the 405 that wasn't — speaking pip Manager's legacy-UI dialect](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-18.md): Custom-node and model operations no longer 405 against a pip ComfyUI-Manager running in legacy-UI mode (hardcoded by popular Docker images): comfyui-mcp now detects the mode and speaks its queue/batch dialect, and Manager detection can no longer be fooled by an SPA's 200. - [July 17: Kimi K3, the CivitAI browser grows up, panel-owned sessions](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-17.md): Moonshot's Kimi K3 becomes a first-class provider; panel 0.9.0 turns the CivitAI browser into a full loop — browse, like, filter by creator, and load embedded workflows (even from zip wrappers) straight onto the canvas — makes panel-owned sessions the default so conversations survive workflow switch… - [July 16: live queue status everywhere, one-tap mobile cancel](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-16.md): The orchestrator broadcasts ComfyUI's live render/queue state to every connected tab (powering the mobile queue monitor), cancel_job joins the mobile whitelist for one-tap cancel, list_local_models links models with CivitAI provenance sidecars back to their CivitAI pages, and the watchdog WebSocket… - [July 15: flatten workflows, CivitAI creator discovery, desktop-tab mirror](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-15.md): panel_flatten_workflow resolves Get/Set buses, Reroutes, and use-everywhere links in place without moving a single kept node; search_civitai_creators + a creator filter make 'top creators → their models → download' work end-to-end; and the phone can mirror and drive a desktop tab's session. - [July 14: the self-updating orchestrator, vision for every backend](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-14.md): The panel orchestrator now updates and restarts itself when every agent is idle — a running install never goes stale. Every Ollama-family backend and ChatGPT-direct gains inline image delivery with an honest fallback when a model is text-only, the gemma4 fine-tunes get correct sampling, and a stable… - [July 12: five new providers, A2UI chat cards, per-workflow sessions](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-12.md): An OAuth engine brings Grok, Kimi, GLM, ChatGPT-direct, and Copilot as panel providers; validated interactive A2UI cards render in chat; each workflow can get its own agent session; a loopback MCP console and credential slots land — plus native CivitAI search, official comfy-cli integration, and the… - [July 9: LM Studio & llama.cpp backends, phone pairing, gemma4 Arena results](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-09.md): Every local runtime becomes a first-class backend (LM Studio, llama.cpp, custom OpenAI-compatible endpoints), a token-gated phone-pairing listener and render mailbox land for mobile, graph_query filters big workflows, and the gemma4 fine-tune ladder posts its Arena scores. The panel keeps working wh… - [July 8: local LLMs, remote pods, and the RunPod image — 23 releases in one day](https://comfyui-mcp.artokun.io/docs/changelog/2026-07-08.md): The agent goes all-LLM: an Ollama backend with a compact tool mode and our gemma4-comfyui-mcp fine-tune ladder as the local default, benchmarked in a new ComfyUI LLM Arena. Plus one-command connect to a remote ComfyUI over a secure wss:// bridge, a custom RunPod image, and speaking both ComfyUI-Mana… - [June 29: Comfy MCP parity, run-to-node, graph navigation](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-29.md): Five one-call generation tools close the gap with Comfy's official cloud MCP (plus a token-gated tunnel connector), panel_run learns to render a single output branch, and the panel agent gains graph outline/search/subgraph tools plus awareness of your manual canvas edits. - [June 27: render-wedge watchdog, color analysis, provider onboarding](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-27.md): A self-healing queue/render watchdog (backpressure + stall detection + escalating cancel), a new analyze_color tool, registry-first custom-node installs with a hardened clone fallback, and a wave of panel polish — provider onboarding, rich media metadata, code-block copy/wrap, and a background-tab s… - [June 26: strip_workflow & slice_workflow — load any expert graph](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-26.md): Two new workflow tools resolve GetNode/SetNode buses + Reroutes into real connections and un-chunk rgthree Fast-Groups-Bypass-toggled pipelines — so an agent can load an ad-hoc expert workflow from any path and get clean, runnable wiring without hand-tracing. - [June 25: Multi-provider agent — Claude + ChatGPT](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-25.md): The ComfyUI Agent Panel now runs on EITHER Claude or ChatGPT/Codex at full feature parity — pick a provider, drive your live canvas, load packs in one shot, and stay cost-aware about paid API nodes. - [June 24: Krea 2 Turbo pack](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-24.md): Day-one ComfyUI pack for Krea 2 Turbo — Krea.ai's new 12B open-weight text-to-image model. 8-step turbo speed, Qwen3-VL prompting, JSON area prompts, one-command install. - [June 19: v0.15.0 — streaming panel, SDK commands, one-click recovery](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-19.md): The ComfyUI Agent Panel gains live-streaming chat, SDK slash commands (/compact, /loop, …), one-click recovery for a wedged agent, and pid-reuse-safe process handling. The legacy --channels mode is gone; the workflow converter got tougher and more packs are render-verified. - [June 17: v0.14.0 — the panel drives itself](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-17.md): The ComfyUI sidebar panel now runs an autonomous background agent on your Claude subscription — no API keys, no interactive session needed. Plus a zod 4 migration, a green Windows test suite, and the plugin now shipping in the npm package. - [Week of June 8: v0.9.5 → v0.13.0](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-16.md): Channels mode — drive ComfyUI's canvas from your own Claude session with no API keys — plus the sidebar panel shipping to the Comfy Registry, audio generation, Civitai MCP pairing, and a wave of community reliability work. - [Week of June 1: v0.8.1 → v0.9.4](https://comfyui-mcp.artokun.io/docs/changelog/2026-06-05.md): Comfy Cloud + explicit remote mode, embedded agent sidebar, health checks, slimmer installs, and a stack of stability fixes. - [Week of May 25: v0.6.0 → v0.8.0](https://comfyui-mcp.artokun.io/docs/changelog/2026-05-29.md): Media uploads, hosted docs, model-download hardening, custom-node authoring + verification, declarative setup, image conversion, and cloud storage I/O. ## Other - [Finetune v2](https://comfyui-mcp.artokun.io/docs/design/finetune-v2.md) - [Mobile tunnel reattach](https://comfyui-mcp.artokun.io/docs/design/mobile-tunnel-reattach.md) - [ComfyUI MCP](https://comfyui-mcp.artokun.io/docs/fa/index.md): سطح کنترلِ محلی‌محور و بومیِ عامل برای ComfyUI — یک سرور MCP به‌همراه افزونهٔ Claude Code که با زبان طبیعی تصویر، ویدیو و صدا تولید می‌کند، گردش‌کار می‌نویسد و اجرا می‌کند، و مدل‌ها و گره‌های سفارشی را مدیریت می‌کند. محلی، از راه دور، یا Comfy Cloud. - [نصب](https://comfyui-mcp.artokun.io/docs/fa/installation.md): ComfyUI را نصب کنید و سرور MCP را وصل کنید — محلی، از راه دور، یا Comfy Cloud. - [پنل نوار کناری](https://comfyui-mcp.artokun.io/docs/fa/panel.md): پنل عامل ComfyUI — یک عامل هوش مصنوعی خودگردان در نوار کناری ComfyUI که بوم شما را با هر مدل زبانی هدایت می‌کند: Claude، ChatGPT یا Gemini روی اشتراک خودتان (بدون کلید API)، یک مدل محلی رایگان از راه Ollama (بدون هیچ حسابی)، یا هر مدل میزبانی‌شده روی یک نقطهٔ اتصال سازگار با OpenAI. در Comfy Registr… - [شروع سریع](https://comfyui-mcp.artokun.io/docs/fa/quickstart.md): ComfyUI MCP را به Claude Code اضافه کنید و اولین تصویرتان را تولید کنید. - [رفع اشکال](https://comfyui-mcp.artokun.io/docs/fa/troubleshooting.md): راه‌حل مشکل‌هایی که کاربران واقعاً با آن‌ها روبه‌رو می‌شوند: ناسازگاری نسل‌های ComfyUI-Manager (خطای 405)، نصب از نشانی git که بی‌صدا رد می‌شود، در دسترس نبودن پنل از مرورگر راه دور، کش کهنهٔ npx، و ComfyUI راه دورِ پورت‌فورواردشده که اشتباهاً محلی تشخیص داده می‌شود. ## OpenAPI Specs - [openapi](/docs/api-reference/openapi.json) > The links below point to documentation indexes. Follow each `/_llms/` index recursively until you reach documentation pages. ## Indexes - [Korean (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/ko.md): Documentation for Korean. - [Japanese (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/ja.md): Documentation for Japanese. - [Chinese (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/zh.md): Documentation for Chinese. - [French (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/fr.md): Documentation for French. - [Spanish (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/es.md): Documentation for Spanish. - [Traditional Chinese (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/zh-tw.md): Documentation for Traditional Chinese. - [Russian (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/ru.md): Documentation for Russian. - [Arabic (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/ar.md): Documentation for Arabic. - [Brazilian Portuguese (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/pt-br.md): Documentation for Brazilian Portuguese. - [Turkish (21 pages)](https://comfyui-mcp.artokun.io/docs/_llms/tr.md): Documentation for Turkish.