This server has 4 tools for querying Oura health data. All tools follow a consistent naming pattern (get_*) and have basic descriptions, but critical gaps in schema documentation, parameter descriptions, and output documentation significantly limit LLM usability. Parameters lack type information in descriptions and no output schemas are documented. The server implements basic data transformation but provides no error recovery guidance. Conservative scoring reflects these production-readiness gaps.

• Add output schema documentation to all tools. Each should explicitly list returned fields, their types, and brief descriptions. Example: 'Returns a dict with keys: day (string), bedtime_start (string, ISO 8601), total_sleep_duration (string, human-readable format), average_heart_rate (integer), readiness_score (integer, 0-100 or null).'
• Enhance tool descriptions to 100-200 characters. Explain semantic differences: 'Get detailed sleep session data including bedtime, duration by stage (deep/light/REM), heart rate, and HRV for a date range. Use get_daily_sleep_data for aggregated daily scores instead.'
• Update parameter descriptions to specify format and constraints: 'start_date (required): ISO 8601 date string (YYYY-MM-DD). Must be in the past or today. Example: 2024-01-15' and 'end_date (optional): ISO 8601 date string. If omitted, defaults to start_date. Must be >= start_date.'
• Move 'defaults to start_date' from description into the JSON schema as a default value, not free text.
• Add error recovery hints to descriptions, e.g., 'If you receive an authentication error, verify the OURA_ACCESS_TOKEN environment variable is set correctly. If a date range returns no data, try a more recent range or call search_available_dates() first.'
• Document returned fields as enum or structured example. This helps LLMs understand what data is available without guessing.
• Consider splitting get_sleep_data and get_daily_sleep_data into separate, clearly named variants if they serve truly different use cases, or merge them with a parameter flag if they share implementation.