Well-structured expense tracker with solid schema definitions and clear naming. All 5 tools have explicit descriptions and input schemas with Zod validation. Tool names follow verb_noun pattern (add_, update_, delete_, list_, get_). Schemas are comprehensive with format validation, range constraints, and helpful error messages. However, output schemas are partially documented (only get_spending_summary has explicit output schema), and descriptions are brief (10-30 chars), below the 50-200 char LLM-optimized baseline. Error handling is present but lacks recovery guidance. No tool annotations visible in registration code.

• Expand tool descriptions to 50-200 characters. Example: 'add_expense' → 'Record a new expense with amount, category, and date. Use this to log spending for budgeting and analysis. Returns the expense ID for future reference.'
• Document output schemas for all tools. Define what fields add_expense returns (e.g., {id, amount, category, date, description, createdAt}), and ensure list_expenses returns {expenses: [...], total, hasMore}.
• Add tool annotations in server.ts registration: mark delete_expense with destructiveHint=true, add_expense and update_expense with idempotentHint=false (since they modify state), and list_expense/get_spending_summary with readOnlyHint=true.
• Enhance error messages with actionable guidance. Wrap Zod validation errors to return: 'Invalid amount: must be a positive number ≤ 1,000,000. Got: -50' instead of raw validation failure.
• Add pagination to list_expenses: accept optional limit (1-100, default 20) and offset parameters; return {expenses, total, offset, limit, hasMore} so agents can iterate through large expense histories.