Loading ToolBench
Remote control bridge for ComfyUI — server extension, Python client, and MCP server. Bridges LLM tool-use (Claude Code, Cursor, etc.) to ComfyUI's web frontend via the uiapi extension.
11 tools with mixed quality. Naming is verb-forward and clear (generate_image, check_status, list_routes, query_fields, set_fields, set_connection, execute, get_workflow, get_workflow_api, show_workflow_dialog, get_model_url). Descriptions range 50 - 250 chars, mostly adequate but some lack actionable context. Input schemas are present and typed for all tools, but parameter descriptions vary in specificity. Critical gap: no output schemas documented anywhere, LLMs cannot infer what these tools return. Error handling is absent; no recovery guidance. Tool composition is reasonable (single responsibility), but several tools require 'active WebUI browser tab' which is a hidden dependency not enforced at the API level.
Check the status of a ComfyUI workflow execution by prompt ID. Returns execution status and any available outputs.
Execute the current ComfyUI workflow. Converts the browser-loaded workflow to a prompt and queues it for execution. Requires an active WebUI browser tab.
Generate an image using an API provider (Gemini, OpenAI, Grok, Flux, Stability, Ideogram, Kling, Luma, etc.). Supports text-to-image and image-to-image workflows. Automatically discovers available providers from ComfyUI's /object_info.
Request download URLs for one or more model checkpoints. Displays a dialog in the ComfyUI browser tab for the user to enter download URLs. Requires an active WebUI browser tab.
Retrieve the current ComfyUI workflow from the browser as a complete workflow object. Requires an active WebUI browser tab.
Retrieve the API representation of the current ComfyUI workflow from the browser. Returns only the 'output' section suitable for API calls. Requires an active WebUI browser tab.
No output schemas documented for any tool. LLMs cannot infer return types, field names, or structure. Breaks downstream tool chaining and forces agents to guess what data is available.
Hidden dependency: 'Requires an active WebUI browser tab' stated in descriptions but not enforced via schema or error handling. LLMs cannot know which tools need this precondition without reading descriptions carefully.
No error handling or recovery guidance. Tools fail silently or with generic errors. No actionable messages to guide LLM retry or fallback strategies.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | <=2025-11-25 | v2 |
List all available API image generation providers discovered from ComfyUI's /object_info. Shows provider names, available models, and capabilities.
Query the current ComfyUI workflow in the browser. Returns all nodes, their types, titles, inputs, and outputs. Requires an active WebUI browser tab.
Create a connection between two nodes in the ComfyUI workflow. Connects the first output of source_node to the first input of target_node. Requires an active WebUI browser tab.
Set widget values in the current ComfyUI workflow. Accepts a list of [path, value] pairs where path is 'node_id.widget_name'. Requires an active WebUI browser tab.
Display a workflow dialog in the ComfyUI browser tab asking the user to accept or reject a workflow. If accepted, loads the workflow into the browser. Requires an active WebUI browser tab.
Parameter descriptions lack specificity. 'Enable verbose logging' is vague; 'Aspect ratio or size preset' does not enumerate valid values. LLMs cannot self-correct invalid inputs.
Tool composition: generate_image, check_status, and list_routes form a multi-step workflow but lack guidance on sequencing. Instructions mention 'do NOT call check_status first' but this should be enforced via tool design, not prose.