Let an AI inspect, edit, and queue the workflow currently open on your ComfyUI canvas.
Strong schema definitions with comprehensive parameter constraints and clear descriptions. All 9 tools have explicit input schemas with proper type definitions, enums, and validation rules. Descriptions are generally 50-150 chars, meeting the 10-1024 baseline. Tool names follow verb_noun convention (list_, inspect_, search_, apply_, queue_, wait_). However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Error handling guidance is absent; tools do not indicate what to do on failure. Some parameter descriptions lack actionable format hints (e.g., 'canvas_id' is opaque with no guidance on how to obtain it). Composition is strong: tools chain well (list_instances → inspect_canvas → apply_canvas_patch), and most operations are idempotent or clearly stateful.
Apply one atomic, undoable batch of changes to a live ComfyUI canvas.
Inspect a compact live ComfyUI canvas, subgraph navigation and ports, or details for native node and group refs.
Inspect the complete native ComfyUI schema for one installed node type.
Inspect the status and outputs of a queued prompt.
Discover local ComfyUI instances, connection status, last focus times, and their active canvases.
Navigate to a native graph, select and optionally fit its items without modifying the workflow.
Queue a prompt for execution on ComfyUI.
Output schemas not documented. LLMs cannot predict response structure, breaking tool chaining and forcing exploratory calls.
No error handling guidance. Tools do not indicate recovery paths (e.g., 'If canvas_id is invalid, call list_instances first').
Opaque identifiers (canvas_id, instance_id, prompt_id) lack discovery guidance. Descriptions do not explain how to obtain these IDs or what they represent.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Search the node types installed in ComfyUI.
Wait for a queued prompt to complete execution.
apply_canvas_patch operations array is complex with 15 oneOf variants. No guidance on operation ordering, atomicity guarantees, or partial failure behavior.
Parameter descriptions for 'refs' array in inspect_canvas and present_canvas do not explain the structure or how to construct valid ref objects.