RunComfy connector
Headless control of RunComfy dedicated ComfyUI machines (pods) via the Server API, plus queueing workflows on a pod’s live ComfyUI backend.Environment
MCP tools
runcomfy_list_pods—GET /prod/api/users/:user/servers. Lists pods withserver_id,current_status,main_service_url,workflow_version_id.runcomfy_sync_workflow—GET /prod/api/users/:user/workflows. Lists cloud workflows andworkflow_version_idvalues. Optionallocal_workflow_pathname-matches a local JSON file to a cloud workflow.runcomfy_queue— POST workflow to a pod’smain_service_url/prompt. Useserver_idormain_service_urlfor an existing pod, orworkflow_version_idto launch then queue (polls until ComfyUI responds).
Queue flow
runcomfy_sync_workflow→ picklatest_version_id/matched_version_idruncomfy_queuewithworkflow_version_id+workfloworworkflow_path(launches pod if needed)- Or
runcomfy_list_pods→ reuseserver_idwhenmain_service_urlis ready
convertUiToApi using the pod’s remote /object_info.
Serverless API
RunComfy also offers serverless deployments athttps://api.runcomfy.net (/prod/v2/deployments/.../inference). This phase targets dedicated pods (Server API) per the roadmap; serverless can be a follow-up.
Related
- Phase 3: ai-toolkit-supervisor.md (local training UI)
- agent-platform-roadmap.md