The server defines 2 tools with visible schemas and reasonable descriptions. Both tools follow a clear naming pattern (verb_noun: whoop_sleep_recent, whoop_cycle_strain) and have documented input parameters with types and constraints. However, there are significant gaps: (1) parameter descriptions are minimal or absent in the actual schema definitions visible in src/app.ts, the Zod schemas define types and constraints but lack the descriptive text an LLM needs to understand when and how to use each parameter; (2) output schemas are not formally documented, responses include raw API payloads with a 'structuredContent' field, but the schema of that content is not defined in the tool registration; (3) no error handling guidance is present to tell the LLM how to recover from failures; (4) security considerations around token storage and multi-user key support are not explained in tool descriptions. The tools are read-only and relatively safe, but the lack of parameter descriptions and output schema documentation represents a material gap.

• Convert Zod schemas to explicit JSON Schema objects with 'description' fields for each parameter. Example: instead of z.number().int().min(1).max(25), use {'type': 'integer', 'minimum': 1, 'maximum': 25, 'description': 'Maximum number of sleep sessions to fetch per request. Default: 20.'}
• Document the output schema in the tool registration. Define the structure of 'structuredContent' returned by each tool, including field names, types, and whether arrays include pagination info.
• Add error recovery guidance to error messages. Instead of 'Failed to fetch sleep data: <error>', return 'Failed to fetch sleep data: <error>. This usually means the WHOOP API is unavailable or your authentication token has expired. Try again in a moment or re-authenticate if needed.'
• Add description to the 'key' parameter explaining that it selects which user's token to use for multi-user setups. Example: 'Token store key for multi-user support. Use different keys to fetch data for different WHOOP accounts. Defaults to "default" (the current user).'
• Document date parameter format explicitly. Add to 'start' and 'end' descriptions: 'ISO 8601 date-time string (e.g., "2024-01-15T08:30:00Z"). If omitted, defaults to the last 30 days.'
• Clarify pagination. Update 'nextToken' description: 'Pagination token from a previous response. Pass the exact value returned in next_token to fetch the next page of results. Omit this parameter for the first request.'
• Consider reducing verbosity of responses. The inclusion of raw JSON payloads in every response ('Raw payload: {...}') inflates token cost. Return only the summarized text content and structured fields the LLM needs.

Inferred effective spec: 2026-07-28+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained