MCP server providing ground-truth Unity API documentation (2022/2023/6) -- prevents AI hallucination of signatures, namespaces, and deprecated APIs
The server provides 5 well-intentioned tools for Unity API documentation lookup. All tools have explicit descriptions and are named with clear action verbs (search_, get_). However, there are significant gaps in schema documentation, parameter descriptions lack actionable constraints, and output schemas are not formally documented. The tools are read-only and safe, but the descriptions could be more directive for LLM selection. No error handling guidance is provided, tools return plain text with fallback behavior but don't structure error responses for agent recovery.
Get a complete reference card for a Unity class — all public methods, properties, fields, and events. Use this when you want to see everything available on a class at a glance.
Check whether a Unity class, method, or property is deprecated, and find the replacement. Use this before using any API you suspect might be outdated. Catches common mistakes like using FindObjectOfType (deprecated) instead of FindAnyObjectByType.
Look up the exact signature of a Unity API member by fully-qualified name. Use this when you know the specific API (e.g. from autocomplete or docs reference) and need its exact parameters, return type, and deprecation status.
Find the correct namespace and using directive for a Unity class or member. Use this when you need to know which namespace to import for a Unity type.
Search Unity API documentation by keyword. Use this to find Unity classes, methods, properties, and fields. Returns ranked results with summaries.
Output schemas are not formally documented. All 5 tools return plain text formatted with _format_record(). Agents cannot parse structured data without manual text extraction, wasting tokens and increasing errors.
Parameter 'member_type' in search_unity_api lists allowed values in description ('type', 'method', 'property', 'field', or 'event') but not as a JSON Schema enum constraint. LLMs infer valid values from text and may hallucinate invalid options like 'class' or 'interface'.
No error handling or recovery guidance. Tools return fallback results (FTS search, did-you-mean) but descriptions don't mention these behaviors. If a tool returns no results, agent doesn't know whether to retry, relax constraints, or ask the user.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 62 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 54 | - | v1 |
Tools return all available results formatted as text. search_unity_api caps at 20 results (n_results parameter) but others have no limit. Large result sets waste tokens, no pagination or result-count guidance in descriptions.
Descriptions use example values ('Physics Raycast', 'FindObjectOfType') rather than formal constraints or enums. LLMs may reuse examples literally instead of adapting to actual queries.