Healthcare MCP server with 7 tools targeting women's health tracking (menstrual cycles, hormone tests, daily health records). Tool naming is consistently verb-based (get_, record_) and clear. All tools have descriptions (100-200 chars, within baseline). Input schemas are present and typed for all tools with proper enum constraints (e.g., symptom codes, test types, hormone types). Output schemas are NOT documented in the source code, the server definition shows input schemas but no return type documentation. Parameter descriptions are present but vary in quality; some are minimal (e.g., 'Array of symptom codes' lacks context on when/why to use). Error handling is not visible in the source excerpt. Security: no exposed credentials in tool definitions; OAuth provider present suggests proper auth gating. Tool composition is reasonable, separate read/write tools for each domain (daily health, menstrual, hormone). No evidence of idempotency markers or confirmation steps for destructive operations.

• Add output schema documentation to every tool. For record_* tools, return { success: boolean, id: string, createdAt: ISO8601, ... }. For get_* tools, return { items: [...], total: number, nextCursor?: string, ... }. This enables tool chaining and error handling.
• Implement pagination for list tools. Add limit (default 20, max 100) and cursor/offset parameters. Update descriptions to state result limits.
• Document idempotency semantics for record_* tools. Specify which fields constitute a duplicate (e.g., same date + same hormone test type) and what happens on retry (merge, reject, or replace).
• Add confirmation/dry-run mode for destructive operations. Consider a 'confirm_before_execute' pattern or separate dry-run tools for record_* operations targeting sensitive health data.
• Enhance parameter descriptions with clinical context. E.g., 'symptomMemo: Additional notes about symptoms (e.g., severity, timing relative to cycle, triggers). Helpful for identifying patterns.' This guides LLM prompting.
• Add explicit error codes and recovery guidance. E.g., 'If hormone level is invalid (must be 0-5), the tool returns { error: "invalid_range", message: "Hormone level 7.2 exceeds maximum of 5", suggestion: "Check the test result label and resubmit." }'. Violates pattern:recovery-guide without this.
• Document field relationships. E.g., 'endDate is optional for record_menstrual_period; if omitted, the system assumes period is ongoing.' Prevents silent misuse.
• Add response field validation. Ensure returned date strings are ISO 8601 (not Unix timestamps or custom formats). Ensure all IDs returned by record_* tools can be passed to get_* filters.