comfyui-api

Installation

$npx skills add MCKRUZ/ComfyUI-Expert --skill comfyui-api

Summary

After loading this skill, the agent can connect to a running ComfyUI instance, validate and queue image generation workflows, poll for execution progress, and retrieve output images. Invoke when the agent needs to execute ComfyUI-based image or video generation tasks.

SKILL.MD

ComfyUI API Skill

Connect to ComfyUI's REST API to execute workflows, monitor progress, and retrieve outputs.

Configuration

  • Default URL: http://127.0.0.1:8188
  • Custom URL: Set in project manifest or pass as parameter
  • Timeout: 30s for API calls, no timeout for generation polling

Two Modes

Online Mode (ComfyUI Running)

Full API access. Preferred mode for interactive work.

  1. Test connection: GET /system_stats
  2. Discover capabilities: Use comfyui-inventory skill
  3. Queue workflow: POST /prompt
  4. Poll for results: GET /history/{prompt_id} every 5 seconds
  5. Retrieve outputs: GET /view?filename=...

Offline Mode (No Server)

Export workflow JSON for manual loading in ComfyUI.

  1. Generate workflow JSON following ComfyUI's format
  2. Save to projects/{project}/workflows/{name}.json
  3. Instruct user to drag-drop into ComfyUI

API Operations

Check Server Status

curl http://127.0.0.1:8188/system_stats

Response fields:

  • system.os: Operating system
  • system.comfyui_version: Version string
  • devices[0].name: GPU name
  • devices[0].vram_total: Total VRAM bytes
  • devices[0].vram_free: Free VRAM bytes

Queue a Workflow

curl -X POST http://127.0.0.1:8188/prompt \
  -H "Content-Type: application/json" \
  -d '{"prompt": WORKFLOW_JSON, "client_id": "video-agent"}'

WORKFLOW_JSON format:

{
  "1": {
    "class_type": "LoadCheckpoint",
    "inputs": {
      "ckpt_name": "flux1-dev.safetensors"
    }
  },
  "2": {
    "class_type": "CLIPTextEncode",
    "inputs": {
      "text": "photorealistic portrait...",
      "clip": ["1", 1]
    }
  }
}

Each node is keyed by a string ID. Inputs reference other nodes as ["{node_id}", {output_index}].

Response:

{"prompt_id": "abc-123-def", "number": 1}

Poll for Completion

curl http://127.0.0.1:8188/history/abc-123-def

Incomplete: Returns {} (empty object) Complete: Returns execution data with outputs:

{
  "abc-123-def": {
    "outputs": {
      "9": {
        "images": [{"filename": "ComfyUI_00001.png", "subfolder": "", "type": "output"}]
      }
    },
    "status": {"completed": true}
  }
}

Retrieve Output Image

curl "http://127.0.0.1:8188/view?filename=ComfyUI_00001.png&subfolder=&type=output" -o output.png

Upload Reference Image

curl -X POST http://127.0.0.1:8188/upload/image \
  -F "image=@reference.png" \
  -F "subfolder=input" \
  -F "type=input"

Cancel Current Generation

curl -X POST http://127.0.0.1:8188/interrupt

Free VRAM

curl -X POST http://127.0.0.1:8188/free \
  -H "Content-Type: application/json" \
  -d '{"unload_models": true}'

Polling Strategy

ComfyUI doesn't support WebSocket in CLI context. Use REST polling:

  1. Queue workflow via POST /prompt → get prompt_id
  2. Poll GET /history/{prompt_id} every 5 seconds
  3. On empty response: generation in progress, continue polling
  4. On populated response: check status.completed
  5. If completed: true, extract outputs
  6. If error in status, route to comfyui-troubleshooter

Timeout: Warn user after 10 minutes of polling. Video generation (Wan 14B) can take 15-30 minutes.

Workflow Validation

Before queuing any workflow:

  1. Read state/inventory.json (via comfyui-inventory)
  2. For each node in workflow: verify class_type exists in installed nodes
  3. For each model reference: verify file exists in installed models
  4. Flag missing items with:
    • Node: suggest ComfyUI-Manager install command
    • Model: provide download link from references/models.md
    • Version mismatch: suggest update

Error Handling

ErrorCauseAction
Connection refusedComfyUI not runningSwitch to offline mode, save JSON
400 Bad RequestInvalid workflow JSONValidate node connections
500 Internal ErrorComfyUI crashSuggest restart, check logs
Timeout (no response)Server overloadedWait and retry once

Reference

Full API documentation: foundation/api-quick-ref.md