Loading ToolBench
MCP server for ComfyUI Workflow Architect; exposes ComfyUI operations as tools for any MCP-compatible AI client
15 tools with inconsistent quality. Most have basic descriptions (50-150 chars) but lack parameter-level documentation. Input schemas visible for all tools but many parameters undescribed. No output schemas documented. Error handling absent, tools return raw API responses without recovery guidance. Naming is verb-first and clear (get_, list_, prompt, interrupt), but composition issues exist: prompt + interrupt could be unified; validate_workflow duplicates get_object_info logic. No tool annotations (readOnlyHint/destructiveHint). STDIO transport caps protocol score at 50.
Clear internal caches (object_info, model paths)
List available embeddings in ComfyUI
Fetch execution history from ComfyUI
Get the detailed schema for a specific node type
Fetch the /object_info endpoint from ComfyUI, which contains all available node types and their schemas
Get the current queue status from ComfyUI (running and pending jobs)
Get system information from ComfyUI including VRAM, device info, and configuration
Interrupt the currently running ComfyUI execution
No output schemas documented. LLMs cannot predict response structure, forcing them to guess field names for downstream tool calls. E.g., get_object_info returns object_info but no schema shows field types or nested structure.
Parameter descriptions missing or trivial for 8+ tools. E.g., list_custom_nodes has empty input schema; get_queue_status, interrupt, get_system_info, list_workflows, get_embeddings, clear_caches lack parameter docs. LLMs cannot infer when/how to call these tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
List all installed custom node packs in the ComfyUI custom_nodes directory
List saved workflows from ComfyUI
Load a saved workflow from ComfyUI
Execute a workflow prompt on ComfyUI
Save a workflow to ComfyUI
Scan the ComfyUI models directory for available model files by category
Validate a ComfyUI workflow against the live schema from /object_info
No error handling or recovery guidance. Tools return raw API responses (or failures) without actionable next steps. E.g., if prompt fails, LLM has no guidance on whether to retry, validate workflow first, or check system_info.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe reads from destructive writes. E.g., interrupt and clear_caches are destructive but unmarked; get_* tools are read-only but unmarked.
Composition issue: validate_workflow and get_object_info both fetch schema. prompt + interrupt could be unified into execute_workflow with optional interrupt flag. Redundancy wastes LLM reasoning cycles.