The server has 6 well-named tools with explicit schemas and descriptions present in src/index.ts. Tool names follow verb_noun convention (get_today, get_recovery_trends, sync_data, get_auth_url). Descriptions are present and reasonably specific (avg ~120 chars, baseline 194 chars). However, several parameter descriptions are sparse or missing detail. Input schemas are properly typed with JSON Schema. Major gaps: (1) No documented output schemas, responses are formatted as markdown text rather than structured objects; (2) No parameter descriptions for most numeric parameters (e.g., 'days' parameter lacks guidance on what happens at boundary values); (3) Error handling is minimal, tools return plain text errors instead of actionable recovery guidance; (4) No tool annotations (readOnlyHint, destructiveHint, idempotentHint); (5) Response formatting as text rather than structured JSON limits downstream tool composition. The server is functional but misses LLM-optimization patterns for chainability and structured reasoning.

• Define output schemas for all tools. Instead of returning markdown text, return structured JSON: { recovery_score: number, hrv_rmssd: number, resting_hr: number, ... }. This enables downstream tool chaining and LLM-structured reasoning.
• Expand 'days' parameter description to: 'Number of days of historical data to include (1 - 90; default 14). Range is inclusive of today. Minimum 1 returns only the most recent complete day.'
• Replace plain-text error responses with structured error objects: { error: 'not_authenticated', code: 'AUTH_REQUIRED', message: '...', recovery: 'Call get_auth_url() to authorize...' }. Categorize as 'retryable', 'user_action_required', or 'fatal'.
• Add tool annotations to ListToolsRequestSchema response: mark all get_* tools with readOnlyHint=true, mark sync_data with destructiveHint=true, mark get_today (likely idempotent on repeated calls) with idempotentHint=true.
• Document the OAuth callback flow in get_auth_url description: 'Returns the Whoop authorization URL. After user clicks the link and grants permission, the callback handler (running at your configured redirect_uri) automatically saves tokens. No further manual action required.'
• Add description to 'full' parameter: 'If true, force a full 90-day historical sync even if recent data is cached (slower, more complete). If false (default), perform an incremental sync, skipping days already in local database (faster, suitable for frequent polling).'
• Return a count/limit in responses when tools return multiple items. E.g., get_recovery_trends should return: { count: 14, data: [...], period_days: 14 }. This prevents context explosion if the API someday changes result size.