Loading ToolBench
Local runtime evidence system for AI-assisted Godot development. Verify, run, inspect, screenshot, and playtest Godot 4.x projects from Cursor, Claude Desktop, Codex CLI, or any MCP-compatible client.
16 tools with consistent naming (verb_noun pattern: godot_sight_*) and basic schemas. However, descriptions are minimal (10-15 chars, well below the 50-200 char LLM-optimized baseline), parameter descriptions are generic or missing, and output schemas are entirely undocumented. No error recovery guidance, no parameter constraints (enums, ranges), and no documentation of what each tool returns. Tool names are clear but descriptions fail to explain WHEN to use each tool or what it returns. This is a typical community server with functional definitions but poor LLM optimization.
Run doctor through the Godot Sight supervisor.
Run logs through the Godot Sight supervisor.
Run runtime-state through the Godot Sight supervisor.
Run scene-tree through the Godot Sight supervisor.
Run input through the Godot Sight supervisor.
Run report latest through the Godot Sight supervisor.
Run playwrite capture start through the Godot Sight supervisor.
Descriptions are 10-15 characters, far below the 50-200 char LLM-optimized baseline. Examples: 'Run doctor through the Godot Sight supervisor.' does not explain WHEN to call doctor, what it returns, or what it checks. LLMs cannot determine tool selection from such minimal text.
Output schemas are completely undocumented. No tool documents what fields it returns, what data types are in the response, or what the LLM should expect. This forces LLMs to guess at response structure and breaks downstream tool chaining.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2025-06-18+ | v2 |
Run playwrite capture stop through the Godot Sight supervisor.
Run playwrite promote through the Godot Sight supervisor.
Run playwrite replay through the Godot Sight supervisor.
Run project-info through the Godot Sight supervisor.
Run playtest through the Godot Sight supervisor.
Run run through the Godot Sight supervisor.
Run screenshot through the Godot Sight supervisor.
Run stop through the Godot Sight supervisor.
Run verify through the Godot Sight supervisor.
Parameter descriptions are generic or missing. 'Absolute path to the Godot project directory' appears 11 times but does not explain what happens if the path is invalid, whether relative paths are accepted, or what format is expected. No constraints (min/max, regex, enums) are documented.
No error handling guidance. The code returns failure() with messages like 'Supervisor command failed for godot_sight_run_scene: ...' but does not tell the LLM whether to retry, ask the user, or give up. No recovery suggestions or actionable next steps.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). The risk field in the spec (READ_ONLY vs WRITE) is not exposed to the MCP protocol, so clients cannot determine which tools are safe to call without user confirmation.