Two tools with basic schemas and descriptions, but significant gaps in parameter documentation, error handling, and output schema clarity. Tool names are action-oriented (generate_text, transcribe_audio) but descriptions lack LLM-optimized structure. Parameters have types and basic descriptions, but missing constraints (enums, ranges, format specs). No documented output schemas. Error handling is minimal, no recovery guidance or actionable error messages. The server accepts base64 audio and model names but lacks validation rules, format constraints, and dependency documentation. Descriptions are in Spanish, which may confuse English-speaking LLMs.

• Document output schemas for both tools. E.g., generate_text returns {text: string, model: string, usage: {prompt_tokens: int, completion_tokens: int}}. transcribe_audio returns {text: string, language: string, duration_seconds: float}.
• Add enum constraints for 'model' parameters. generate_text: ['gpt-4', 'gpt-4-turbo', 'gpt-3.5-turbo']. transcribe_audio: ['whisper-1']. Prevent hallucinated model names.
• Enhance error messages with recovery guidance. E.g., 'Invalid model: got "gpt-5". Valid models: gpt-4, gpt-4-turbo, gpt-3.5-turbo. Call list_models() to see all available options.' instead of generic errors.
• Translate tool descriptions and parameter docs to English. LLMs trained primarily on English text perform better with English descriptions, even for non-English APIs.
• Add parameter constraints: audio_content_base64 max length (e.g., 25MB), language as ISO 639-1 enum (e.g., 'en', 'es', 'fr'), and document ffmpeg dependency with fallback behavior.

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

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