ComfyUI MCP Server MCP — Use Cases, Install & Live Demo

Why use it

Key features

Generation: images, audio, video via local ComfyUI
Workflows: list and run saved ComfyUI workflows
Job lifecycle: queue status, get/cancel job
Assets: list, view, get metadata, publish
Config: list models, get/set defaults

Live Demo

What it looks like in practice

comfyui-mcp-server.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "comfyui-mcp-server": {
      "command": "python",
      "args": [
        "server.py"
      ],
      "_inferred": false,
      "env": {
        "COMFYUI_URL": "http://127.0.0.1:8188"
      }
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "comfyui-mcp-server": {
      "command": "python",
      "args": [
        "server.py"
      ],
      "_inferred": false,
      "env": {
        "COMFYUI_URL": "http://127.0.0.1:8188"
      }
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "comfyui-mcp-server": {
      "command": "python",
      "args": [
        "server.py"
      ],
      "_inferred": false,
      "env": {
        "COMFYUI_URL": "http://127.0.0.1:8188"
      }
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "comfyui-mcp-server": {
      "command": "python",
      "args": [
        "server.py"
      ],
      "_inferred": false,
      "env": {
        "COMFYUI_URL": "http://127.0.0.1:8188"
      }
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "comfyui-mcp-server",
      "command": "python",
      "args": [
        "server.py"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "comfyui-mcp-server": {
      "command": {
        "path": "python",
        "args": [
          "server.py"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add comfyui-mcp-server -- python server.py

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use ComfyUI MCP Server

Iterate an image concept with text prompts

👤 Designers and artists using ComfyUI locally ⏱ ~20 min intermediate

When to use: You have ComfyUI set up but want to iterate via chat instead of node graphs.

Prerequisites

ComfyUI running on port 8188 — Follow ComfyUI install; ensure a checkpoint model is in models/checkpoints/
MCP server started — python server.py (listens at http://127.0.0.1:9000/mcp)

Flow

Generate v1

Generate an image: photoreal portrait, overcast light, 1024×1280.✓ Copied

→ Job submitted; id returned; image appears when done
Regenerate with tweaks

Regenerate from that seed — slightly warmer tones, keep composition.✓ Copied

→ Same composition, warmer colors
View + publish

View the latest image; publish it to the gallery as 'portrait-final.png'.✓ Copied

→ Asset metadata + publish URL

Outcome: Fast iteration without clicking nodes.

Pitfalls

Forgetting ComfyUI must be running before the MCP server starts — Start ComfyUI first, then the MCP server

Combine with: filesystem

Kick off a saved ComfyUI workflow from chat

👤 Teams with established ComfyUI workflows ⏱ ~15 min intermediate

When to use: You already built a workflow; now you want to run it programmatically.

Prerequisites

A workflow saved in ComfyUI — Save As → workflow JSON

Flow

List workflows

List my workflows.✓ Copied

→ Array with names/ids
Run it

Run 'product-shot-v3' with positive='ceramic mug on marble', negative='blurry'.✓ Copied

→ Job created; parameters echoed
Wait + grab the result

Tell me when it finishes and show the image.✓ Copied

→ Polls queue; returns asset when complete

Outcome: Workflow orchestration from chat — ideal for batch runs.

Pitfalls

Workflow references custom nodes not installed — Install the required custom nodes in ComfyUI before running

Combine with: comfy-pilot

Turn a short script into a video with a video workflow

👤 Content creators producing shorts ⏱ ~45 min advanced

When to use: You have a ComfyUI video workflow (e.g., AnimateDiff / Mochi) and want to feed prompts sequentially.

Flow

Load the video workflow

Run the 'shorts-v2' workflow with a scene list: [scene prompts].✓ Copied

→ Segments queued; progress per scene
Stitch and publish

Publish the final video.✓ Copied

→ Output path + publish metadata

Outcome: Short videos via local compute, no cloud fees per render.

Pitfalls

GPU memory exhaustion on longer clips — Reduce resolution or frames per segment; batch sequentially

Combinations

Pair with other MCPs for X10 leverage

comfyui-mcp-server + comfy-pilot

Pair this MCP with the Claude-Code-in-ComfyUI editor for full-pipeline authoring

Build the workflow via comfy-pilot, run it via this MCP server.✓ Copied

comfyui-mcp-server + filesystem

Organize generated assets under project folders

After each batch, move outputs into ./assets/<project>/ with dated filenames.✓ Copied

comfyui-mcp-server + mcp-image

Use mcp-image for one-off Gemini gens; ComfyUI for your local SD/FLUX pipelines

Draft hero via mcp-image; deliver brand set via the ComfyUI workflow.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
generate_image	prompt, settings	Standard text-to-image	local compute only
generate_song / generate_video	prompt, workflow	Audio/video generation via dedicated workflows	local compute
regenerate	reference job_id, tweaks	Iterate off a prior result	local compute
view_image	asset_id	Show an asset inline	0
get_queue_status / get_job / cancel_job	job_id	Lifecycle management	0
list_assets / get_asset_metadata	filter	Browse generated outputs	0
list_models / get_defaults / set_defaults	optional model name	Switch checkpoints or adjust defaults	0
list_workflows / run_workflow	name + inputs	Run saved workflows	0 to start
publish_asset / get_publish_info	asset_id	Mark a final asset	0

Cost & Limits

What this costs to run

API quota: Local only — bound by your GPU
Tokens per call: Prompt text small; asset bytes large
Monetary: Free — your electricity and hardware
Tip: Use lower steps / smaller resolution for iteration; reserve full settings for the keeper.

Security

Permissions, secrets, blast radius

Credential storage: No credentials for local ComfyUI. If you proxy it to cloud, add auth at the proxy.

Data egress: ComfyUI runs locally; server traffic is 127.0.0.1 by default. Model downloads pull from HuggingFace / CivitAI depending on your workflow.

run_workflow executes whatever nodes the workflow contains — including custom nodes. Review unknown workflows before running.
Publishing can write to locations a workflow specifies; make sure publish paths stay inside your asset directory.

Troubleshooting

Common errors and fixes

Connection refused 127.0.0.1:8188

Start ComfyUI first; confirm the UI loads in a browser.

Verify: curl localhost:8188/system_stats

Workflow fails with missing custom node

Open ComfyUI Manager and install the required custom nodes, then retry.

Verify: Manager → Install Missing Custom Nodes

Jobs queue but never complete

Check ComfyUI's own console for a CUDA OOM or model loading error.

Verify: Inspect the ComfyUI terminal output

Alternatives

ComfyUI MCP Server vs others

Alternative	When to use it instead	Tradeoff
comfy-pilot	You want an editor that lets Claude draft workflows inside ComfyUI itself	Not a pure MCP server; requires ComfyUI node install
mcp-image	You want zero-setup cloud generation via Gemini	Pay per image; less workflow flexibility
everart	You want style-trained cloud models	Different provider

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills