VoiceStudio MCP demonstrates solid definition quality with comprehensive parameter descriptions and well-structured schemas. All 6 tools have clear descriptions (194-500+ chars, well above 20-char minimum). Parameters use Annotated[T, Field(description=...)] pattern consistently, providing per-parameter LLM guidance. Tool names follow verb_noun convention (clone_voice, synthesize_speech, list_voices, etc.). However, output schemas are not explicitly documented in the source, and error handling guidance is minimal. No tool annotations (readOnlyHint/destructiveHint) visible despite tools having clear risk profiles (WRITE, DESTRUCTIVE). Parameter validation constraints are present (ge/le bounds on numeric params) but some enum-like parameters (stereo_pan, reverb) use string descriptions instead of formal enum constraints, inviting hallucinated values.

• Add explicit output schema documentation for all tools. E.g., 'Returns: {name: str, description: str, language: str, source: str, ref_audio_path: str, created_at: ISO8601}' for get_voice_info.
• Convert stereo_pan and reverb parameters to formal JSON Schema enums instead of free-form strings. This prevents LLM hallucination and makes valid options machine-parseable.
• Add tool annotations (readOnlyHint, destructiveHint, idempotentHint) to FastMCP tool registration. Mark delete_voice as destructive, list_voices and get_voice_info as readOnly, and synthesize_speech as non-idempotent (generates new audio each call).
• Enhance error handling with recovery guidance. E.g., 'If clone fails with "transcript mismatch", verify ref_text matches the audio exactly, including punctuation. Retry with corrected text.'
• Add a note in clone_voice_from_audio description: 'Exactly one of ref_audio_path or audio_url must be provided; passing both or neither will fail.' This clarifies the mutual exclusivity upfront.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry