Claude can operate ComfyUI through the Model Context Protocol (MCP), but you do not have to build a custom server for the basic local workflow: ComfyUI now documents an official comfy-mcp server. It exposes tools for checking a local server, running workflows, finding models and nodes, and retrieving outputs. The named first-person article’s body is not accessible, so its particular implementation and results cannot be verified; the steps below follow ComfyUI’s current official guidance instead.
What MCP does between Claude and ComfyUI
MCP is an open-source standard for connecting AI applications to external systems. In this setup, Claude is the client and an MCP server makes ComfyUI capabilities available as tools Claude can discover and invoke. ComfyUI describes its integration as letting agents search models, nodes, and templates, then run workflows from natural-language requests. MCP documentation · ComfyUI documentation
As an Amazon Associate I earn from qualifying purchases.
The practical sequence is straightforward: you describe a task, Claude chooses an exposed tool, ComfyUI runs the workflow on local or hosted compute, and the result is returned or saved for you. For a local setup, the documented tools include checking server status, submitting a workflow JSON, waiting for a job, and retrieving generated files. The MCP server does not itself start ComfyUI automatically.
Choose local ComfyUI or Comfy Cloud
ComfyUI recommends Cloud for the simplest setup and compatibility with browser-based clients such as claude.ai and Claude Desktop chat. Choose local when you already run ComfyUI on your machine or primarily use a coding agent such as Claude Code.
#1 Best Overall
- Powered by Radeon RX 9070 XT
- WINDFORCE Cooling System
- Hawk Fan
- Server-grade Thermal Conductive Gel
- RGB Lighting
| Decision | Local Comfy MCP | Comfy Cloud MCP |
|---|---|---|
| Where work runs | Your own computer and GPU. | Comfy Cloud GPUs. |
| Setup | Python, a ComfyUI workspace, comfy-cli, and comfy-mcp. |
Hosted endpoint and account authentication. |
| Models and nodes | Uses models, LoRAs, and custom nodes installed locally. | Uses models and nodes supported by Cloud. |
| Client compatibility | Requires a client that can launch a local stdio subprocess. | Remote HTTP connection works with compatible clients. |
| Cost and output | Local generations use your hardware; partner models may use credits. Outputs are written under the local workspace output folder. | Generation requires an active Comfy Cloud subscription. ComfyUI documentation accessed October 7, 2026, says new users are offered five free runs; the offer may change. Outputs are retrieved using a temporary signed URL and a download command. |
The Cloud endpoint is https://cloud.comfy.org/mcp. The local server runs as a process over stdio and does not have a remote URL. These details and service terms can change; consult ComfyUI’s MCP documentation before setup.
Set up the official local MCP server
The documented requirements are Python 3.10 or later, comfy-cli 1.14.0 or later available on your path, a ComfyUI workspace, and a running ComfyUI instance for execution tools. If you already have a workspace, skip the install step.
-
Install the CLI, install ComfyUI if needed, install the MCP package, and launch ComfyUI:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.Rank #2
GIGABYTE GeForce RTX 5070 Ti Gaming OC 16G Graphics Card, 16GB 256-bit GDDR7, PCIe 5.0, WINDFORCE Cooling System, GV-N507TGAMING OC-16GD Video Card- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
pip install "comfy-cli>=1.14.0" comfy install # skip if a workspace already exists pip install comfy-mcp comfy launch -
Register the server with Claude Code. Replace the example executable path with the absolute path to your own
comfycommand:claude mcp add comfy-mcp -e COMFY_BIN=/path/to/venv/bin/comfy -- comfy-mcpIf Claude Code does not inherit an environment containing
comfy, settingCOMFY_BINto its absolute path lets the server locate it. -
For Claude Desktop, add the server to its
mcpServersconfiguration using the format in ComfyUI’s official guide, then restart Claude Desktop so it reads the updated configuration.Rank #3
ASUS TUF Gaming GeForce RTX™ 5080 16GB GDDR7 OC Edition Graphics Card | PCIe 5.0, HDMI/DP 2.1, 3.6-slot, Military-grade Components, Protective PCB Coating, Axial-tech Fans- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
-
Try a bounded request, such as: “Confirm my local ComfyUI is running, then run the workflow at
~/workflows/txt2img.jsonand show me the image.” The expected flow is a status check, workflow execution, and output retrieval.Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallSpecial offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
For current client configuration details, use ComfyUI’s MCP guide and Claude Code MCP documentation.
What the local server can do
The official local server wraps comfy-cli commands. Its documented tools cover:
Rank #4
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
- Checking ComfyUI status; launching or stopping the local instance.
- Running workflows, polling or watching jobs, and retrieving output files.
- Searching templates, nodes, and local models; inspecting custom nodes in the live installation.
- Validating a workflow before running it.
This is enough for a useful integration without writing protocol plumbing yourself. A custom server would make sense if you need behavior the official tools do not expose, but the accessible version of the named article does not establish what its author built, how it was tested, or what results it achieved. The official tool list is a current reference design, not evidence of that post’s implementation.
Check whether your hardware suits local generation
ComfyUI’s current hardware guide offers these vendor recommendations, not independent benchmark results. Actual fit and speed depend on the model and workflow:
Recommended Free Tools
- 24 GB VRAM or more: ComfyUI says this handles most workloads, including video.
- 8–24 GB VRAM: suitable for image generation; video may be slow or may not fit.
- Under 8 GB VRAM: ComfyUI recommends Cloud.
- Mac and current large open-weight models: ComfyUI recommends Cloud because those models will not run at workable speed on Apple GPU, according to its guide.
These are guidance tiers rather than guarantees for every model. Local execution relies on your own installed models, custom nodes, and hardware; if those constraints are a poor fit, Cloud avoids configuring a local GPU but brings subscription-based generation. See ComfyUI’s hardware and MCP guidance for updates.
Best Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Secure the connection before granting access
An MCP server can expose tools that affect files or execute operations, so treat connecting one as a permissions decision rather than a routine plugin install. Anthropic advises users to verify that they trust each MCP server and notes that servers fetching external content can introduce prompt-injection risk. Claude Code MCP security guidance
- Review the server’s source or provenance before connecting it, especially when its tools can read files or run commands.
- Understand which capabilities you are granting and avoid exposing tools beyond the task you need.
- Keep credentials out of checked-in configuration files.
- Be cautious with instructions or content returned from external sources; tool output is not automatically trustworthy.
What a custom ComfyUI MCP server must handle
If the official server does not fit your needs, a custom implementation has two distinct jobs: MCP protocol integration and reliable ComfyUI workflow execution. A sound design needs to define the following:
- Client transport: decide how the intended Claude client launches or connects to the server. The official local route uses stdio; a browser-based client generally needs a compatible remote connection.
- Discoverable tools and input schemas: expose clear operations with constrained inputs, rather than asking Claude to construct arbitrary commands.
- Workflow submission: translate a validated workflow into a ComfyUI run and report whether it was accepted.
- Job lifecycle: wait, poll, or watch for completion, with useful timeout and failure handling.
- Outputs and errors: return generated files or their locations and make validation, execution, and retrieval errors understandable to the client.
The official local server’s status, workflow, job-watching, validation, and output tools provide a practical checklist for the behavior a custom alternative would need to account for. The named post’s specific code and test outcomes remain unverified because its article body was unavailable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
