Two tools with basic schemas and minimal descriptions. text_to_speech has a 26-char description (below 34-char p10 baseline); save_voice_reference has a 51-char description. Both schemas present but lack depth. No parameter descriptions beyond single-line text. No output schema documentation. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Naming is verb-noun compliant but generic. Missing dependency hints, format constraints, and recovery guidance.

• Expand text_to_speech description to 100-150 chars: 'Convert text to speech using Fish Speech TTS. Optionally clone a voice by providing a reference_id from a previously saved voice reference. Returns audio file path or URL.'
• Expand save_voice_reference description: 'Save a voice reference for voice cloning. Requires an audio file path and its transcribed text. Use this before calling text_to_speech with a reference_id. Returns reference_id for reuse.'
• Document output schemas: text_to_speech should return {type: 'object', properties: {audio_path: {type: 'string'}, duration_ms: {type: 'number'}}}. save_voice_reference should return {type: 'object', properties: {reference_id: {type: 'string'}, status: {type: 'string'}}}.
• Add parameter descriptions: reference_id: 'Optional. ID of a previously saved voice reference (from save_voice_reference). If provided, output will use that voice.' reference_audio: 'Absolute or relative path to audio file (WAV, MP3, or OGG format, <30 seconds).' reference_text: 'Exact transcription of the audio file content.'
• Add error handling: Return structured errors like {isError: true, content: [{type: 'text', text: 'Audio file not found at /path/to/file. Verify the path exists and is readable.'}]} with actionable next steps.