A universal Model Context Protocol server for discovering, editing, running, and observing ComfyUI workflows.
ComfyUI MCP demonstrates solid tool coverage (23 tools) with consistent naming patterns (verb_noun: comfyui_*) and descriptions present for all tools. However, significant gaps exist: parameter descriptions are minimal or absent for many inputs, output schemas are not documented, and error handling lacks recovery guidance. Tool names are clear and action-oriented, but parameter documentation falls short of production standards. The server follows a coherent domain model (workflow management, node discovery, execution control) but lacks the LLM-optimization details (constraints, enums, chaining IDs) that distinguish A-grade tools.
Add a node to a managed workflow, optionally checking it against the live ComfyUI.
Remove pending workflows from the ComfyUI queue.
Connect one node output index to a named input on another node.
Create an empty managed API-format workflow.
Remove the node connection currently assigned to an input.
Download a generated ComfyUI file into this server's managed output directory.
Get the authoritative inputs, outputs, options, and metadata for one node class.
Output schemas not documented. Tools return dict[str, Any] with no specification of fields, types, or structure. LLMs cannot plan downstream calls or extract required data without trial-and-error.
Parameter descriptions are minimal or missing. Many parameters (e.g., 'query', 'folder', 'mode', 'overrides') lack detail on format, constraints, or valid values. LLMs cannot self-correct invalid inputs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Inspect running and pending jobs; full workflow payloads are hidden by default.
Get a run's pending, running, completed, failed, or unknown status and outputs.
Read a managed workflow. Set include_graph=false for a compact summary.
Check ComfyUI connectivity and whether the optional canvas bridge is installed.
Import a ComfyUI API-format workflow object into the managed workflow store.
Interrupt the workflow currently executing in ComfyUI.
List ComfyUI model folders, or models within one folder such as checkpoints or loras.
Search installed core and custom node types without returning the full large schema.
List workflow templates exposed by installed ComfyUI node packages.
List workflows managed by this MCP server with revision and node counts.
Remove a node; force=true also disconnects inputs that reference the node.
Submit a managed workflow to ComfyUI, with optional non-persistent input overrides.
Set a literal node input such as prompt text, seed, dimensions, or model name.
Render a managed API workflow on an open ComfyUI canvas via the optional bridge.
Upload an allowed local image to ComfyUI for LoadImage-style nodes.
Validate graph structure and, by default, installed node schemas and required inputs.
No enum constraints on string parameters. 'mode' in comfyui_sync_canvas accepts 'replace' or 'merge' but is not declared as an enum. 'output_type' in comfyui_download_output lacks enumeration. Free-form strings invite hallucinated values.
Error handling lacks recovery guidance. No evidence of actionable error messages, categorization (retryable vs fatal), or suggestions for next steps. Agents cannot self-correct on failure.
No pagination support visible. comfyui_list_node_types, comfyui_list_models, comfyui_list_workflows accept 'limit' but no offset/cursor or total_count. Large result sets risk context window exhaustion.