Loading ToolBench
MCP server for Godot Engine integration providing tools for scene manipulation, project analysis, asset management, runtime control, and editor integration
Critical gaps across naming, descriptions, and schemas. Most tools (17/22) lack visible input schemas in source code. Descriptions are present but minimal (10-50 chars), below the 50-200 char LLM-optimized baseline. Tool names follow verb_noun convention but lack clarity on mutual exclusivity and dependencies. No evidence of error handling guidance, output schema documentation, or parameter validation rules. Only 5 tools show partial schema visibility (query_project_resources, blackboard_write, blackboard_read, blackboard_patch, blackboard_list_keys, blackboard_task_create). The blackboard task tools (17-20) have no visible input schemas at all. This server is typical of community MCP implementations, functional but not production-ready for agent reasoning.
Clear a path or board in the blackboard
List keys in the blackboard with optional prefix filtering
Apply RFC 6902 JSON patch operations to the blackboard
Read a value from the blackboard
Claim a task in the blackboard
Mark a task as complete in the blackboard
Create a task in the blackboard
17 of 22 tools lack visible input schemas in source code. Tools like project_list_autoloads, project_set_autoload, blackboard_task_claim, blackboard_task_update, blackboard_task_complete have no documented parameter types or constraints.
Descriptions are uniformly minimal (10-50 chars), far below the 50-200 char LLM-optimized baseline. E.g., 'List autoloads configured in the project' (42 chars) lacks context on WHEN to use this tool, WHAT it returns, or HOW to interpret results. LLMs cannot distinguish between similar tools (list_autoloads vs list_input_actions) without richer descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
List tasks in the blackboard
Update a task in the blackboard
Write a value to the blackboard
Get control room dashboard data with project status, session information, and server state
Get a project setting value
List autoloads configured in the project
List input actions configured in the project
Remove an autoload from the project
Remove an input action from the project
Search for symbols in project files
Search for text in project files
Set an autoload in the project
Set an input action in the project
Set a project setting value
Query project resources with filtering and fuzzy search
No error handling guidance visible. Tools like blackboard_clear (DESTRUCTIVE risk) and blackboard_patch (RFC 6902 operations) lack recovery hints. E.g., 'If the path does not exist, try blackboard_list_keys() first' or 'Invalid patch operation, expected RFC 6902 format.' Agents cannot self-correct on failure.
Output schemas not documented. Tools return structured data (e.g., query_project_resources returns resource objects, blackboard_read returns values with metadata) but no schema is visible in source. LLMs cannot plan downstream tool calls without knowing what fields to expect.
Parameter dependencies undocumented. E.g., blackboard_write accepts 'ttl_seconds' and 'expected_updated_at_ms' but no description explains when both are required, when they conflict, or what happens if both are omitted. Agents will guess and fail.