Voice intelligence tools for Claude and AI assistants. Exposes ORAVYS voice analysis capabilities (deepfake detection, personality profiling, voice health, speaker comparison) as MCP tools.
Server exposes 5 voice analysis tools with complete input schemas and reasonable descriptions. All tools follow verb_noun naming (analyze_voice, check_deepfake, get_voice_personality, voice_health_check, compare_speakers). Schemas use proper JSON Schema with oneOf constraints for audio input (URL vs base64). However, descriptions are generic and lack actionable context for LLM selection. No output schemas documented. Missing error handling guidance, parameter validation rules, and recovery paths. No tool annotations (readOnlyHint, destructiveHint). Tier parameter in analyze_voice lacks description of what 'free' vs 'premium' actually returns.
Submit audio for a full ORAVYS voice analysis. Returns emotion detection, personality traits, authenticity score, and cognitive load indicators. Provide either an audio URL or base64-encoded audio data.
Quick deepfake / voice authenticity detection. Uses the ORAVYS proprietary ensemble to determine whether a voice recording is authentic or AI-generated. Provide either an audio URL or base64-encoded audio data.
Compare two voice samples to determine if they belong to the same speaker. Returns a similarity score, confidence level, and same-speaker verdict.
Extract a comprehensive personality profile from a voice sample. Returns trait indicators across multiple validated psychological frameworks.
Assess voice health from an audio sample. Returns voice quality indicators, stress and fatigue levels, an overall health score, and actionable recommendations.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields without knowing response structure (emotion scores, personality traits, deepfake verdict, health metrics, similarity score).
Descriptions lack actionable context. 'Submit audio for a full ORAVYS voice analysis' does not explain WHEN to use this vs check_deepfake or voice_health_check, or what specific use cases it solves. LLMs cannot disambiguate.
analyze_voice 'tier' parameter description is incomplete. States 'free returns core metrics; premium unlocks full engine suite' but does not specify which metrics are core vs premium, or cost/latency implications.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | <=2025-11-25 | v2 |
No error handling guidance. Code raises ValueError and RuntimeError but does not document what errors are retryable, user-fixable, or fatal. LLMs receive raw exceptions with no recovery path.
No tool annotations. Tools are READ_ONLY but lack readOnlyHint in schema. Agents cannot infer safety/idempotency from tool definitions alone.