ComfyUI MCP Server

Instructions

Implementation Reference

• src/comfyui_mcp/server.py:307-314 (registration)

  Registration of the 'list_models' tool using @server.tool decorator.
  @server.tool( name="list_models", description=( "List models exposed by the configured ComfyUI server using the object" " info endpoints. Specify a kind to get a flat list or use recursive" " mode to aggregate multiple kinds." ), )
• src/comfyui_mcp/server.py:315-337 (handler)

  The MCP tool handler for 'list_models', which invokes the model client and logs via context.
  async def list_models( kind: str | None = None, recursive: bool = False, search: str | None = None, context: Context | None = None, ) -> dict[str, Any] | list[str]: """List models available on the remote ComfyUI instance.""" result = await model_client.list_models(kind=kind, recursive=recursive, search=search) if context is not None: if isinstance(result, dict): counts = result.get("counts", {}) await context.info( "Model inventory retrieved", data={"kinds": list(counts.keys()), "counts": counts}, ) else: await context.info( "Model inventory retrieved", data={"kind": kind, "count": len(result)}, ) return result
• src/comfyui_mcp/server.py:157-204 (helper)

  Core logic in ComfyUIModelClient.list_models that queries ComfyUI API object_info for model choices across kinds, applies filtering, and formats the response.
  async def list_models( self, *, kind: str | None = None, recursive: bool = False, search: str | None = None, ) -> dict[str, Any] | list[str]: """Fetch model information from the ComfyUI object info endpoints.""" if self._client is None: raise RuntimeError("Model client not initialised") kinds = [kind] if kind else list(MODEL_KIND_MAP.keys()) if recursive: kinds = list(dict.fromkeys(kinds)) search_lower = search.lower() if search else None inventory: dict[str, list[str]] = {} errors: dict[str, str] = {} for model_kind in kinds: if model_kind not in MODEL_KIND_MAP: raise ValueError(f"Unsupported model kind: {model_kind}") try: choices = await self._choices_for_kind(model_kind) except Exception as exc: # pragma: no cover - defensive logging path inventory[model_kind] = [] errors[model_kind] = str(exc) continue if search_lower: choices = [item for item in choices if search_lower in item.lower()] inventory[model_kind] = sorted(dict.fromkeys(choices)) summary: dict[str, Any] = { "base_url": self.api_base_url, "kinds": kinds, "counts": {kind: len(items) for kind, items in inventory.items()}, "retrieved_at": datetime.now(tz=timezone.utc).isoformat(), "models": inventory, } if errors: summary["errors"] = errors if kind and not recursive: return inventory.get(kind, []) return summary
• src/comfyui_mcp/server.py:205-225 (helper)

  Helper method to fetch model choices for a specific kind by querying the ComfyUI node object_info endpoint.
  async def _choices_for_kind(self, kind: str) -> list[str]: client = self._client if client is None: raise RuntimeError("Model client not initialised") node_class, input_name = MODEL_KIND_MAP[kind] url = f"{self.api_base_url}/object_info/{node_class}" response = await client.get(url) response.raise_for_status() payload = response.json() input_block = payload.get("input", {}) or {} field = ( input_block.get("required", {}).get(input_name) or input_block.get("properties", {}).get(input_name) or {} ) choices = field.get("choices") or field.get("items") or [] if isinstance(choices, dict) and "enum" in choices: choices = choices["enum"] return [item for item in choices if isinstance(item, str)]
• src/comfyui_mcp/server.py:23-29 (helper)

  Mapping from user-facing model kinds to ComfyUI loader node class and input field names.
  MODEL_KIND_MAP: dict[str, tuple[str, str]] = { "checkpoints": ("CheckpointLoaderSimple", "ckpt_name"), "loras": ("LoraLoader", "lora_name"), "vae": ("VAELoader", "vae_name"), "clip": ("CLIPLoader", "clip_name"), "controlnet": ("ControlNetLoader", "control_net_name"), }