Loading ToolBench
Unity Editor integration via the Model Context Protocol (MCP). A bridge that manages and communicates with Unity Editor, allowing MCP clients like Claude Desktop or Cursor to interact with Unity projects.
Significant gaps across all dimensions. Tool names lack clear action verbs (manage_* is vague). Descriptions are present but generic and lack LLM-optimization guidance. Input schemas visible but lack proper type definitions for many parameters (e.g., 'action' is string with enum-like description but no formal enum constraint). Output schemas are not documented. No error handling guidance. Parameters lack descriptions in several tools. The server exposes complex Unity Editor operations without clear recovery paths or validation constraints.
Executes Unity Editor menu items by path. Supports executing menu items and retrieving available menus. Includes a blacklist to prevent execution of potentially disruptive menu items like File/Quit.
Manages prefabs and assets. Supports importing, creating, modifying, deleting, duplicating, moving, renaming, searching, and querying asset information.
Controls editor state and queries info. Supports play mode control (play, pause, stop), editor state queries (get_state, get_windows, get_active_tool, get_selection), tool management (set_active_tool), and tag/layer management (add_tag, remove_tag, get_tags, add_layer, remove_layer, get_layers).
Manages GameObjects in the scene. Supports creating, deleting, finding, modifying properties, managing components, and querying hierarchy information.
Manages scenes. Supports creating, loading, saving, unloading, and querying scene information.
Manages C# script files. Supports creating, modifying, deleting, and querying script information.
Tool names use generic 'manage_*' prefix instead of action verbs. LLMs cannot infer intent from 'manage_editor' vs 'manage_scene', should be 'control_editor', 'load_scene', 'create_scene', etc.
Input schemas lack formal enum constraints. 'action' parameter lists valid values in description ('play, pause, stop, get_state...') but not as JSON Schema enum. LLMs cannot parse description text to extract valid options reliably.
Output schemas not documented. No visibility into what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls or extract required IDs for chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | <=2025-11-25 | v2 |
Reads or clears Unity console messages, with filtering options. Allows retrieving console output with optional filtering by message type and search terms.
No error handling guidance. Tools perform destructive operations (delete, modify, execute menu items) without recovery hints. If 'execute_menu_item' fails, LLM has no guidance on what to try next.
Parameter descriptions incomplete or missing. 'properties' in manage_gameobject and manage_asset is described as 'Properties to set' but does not specify valid keys, types, or constraints. 'waitForCompletion' in manage_editor lacks description entirely.