ExpenseTracker has 4 tools with explicit schema definitions, descriptions, and proper async implementations. However, multiple critical gaps reduce quality significantly. All tools lack parameter-level descriptions in the actual JSON schema (only tool descriptions exist). Output schemas are completely undocumented, callers cannot plan downstream operations. Error messages are generic ('Database error: ...') and provide no recovery guidance. The summarize tool's 'category' parameter defaults to null but lacks clear documentation of what that means. Tool naming follows verb_noun convention (add_expense, list_expenses, delete_expense, summarize) which is good, but descriptions are terse (10-34 chars) and lack WHEN/WHY guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles (WRITE, DESTRUCTIVE). The resource 'categories' is exposed but not referenced by any tool description. No pagination on list_expenses despite potentially unbounded result sets. Overall: adequate for a basic prototype but falls short of production standards.

• Add per-parameter descriptions to every tool schema. For add_expense: 'userID (string): Unique user identifier'; 'date (string): ISO 8601 date, e.g. 2024-01-15'; 'amount (number): Expense amount in dollars, >= 0'; 'category (string): Expense category, see /expense:///categories for valid options'; 'subcategory (string, optional): Subcategory within category'; 'note (string, optional): Free-form note, max 500 chars'.
• Document output schemas explicitly. For add_expense: 'Returns {status: "success" | "error", id: int (new expense ID if success), message: string}'. For list_expenses: 'Returns array of {id: int, date: string (ISO 8601), amount: number, category: string, subcategory: string, note: string}, ordered newest first.'
• Add readOnlyHint to list_expenses and summarize; add destructiveHint to delete_expense. This signals to the LLM which tools are safe to retry and which modify state.
• Expand tool descriptions to 80-150 chars with WHEN/WHY guidance. E.g., add_expense: 'Record a new expense. Call this after the user provides amount, category, and date. Returns the new expense ID for reference.' list_expenses: 'Retrieve all expenses in a date range. Use this to review spending history or before deleting an expense.'
• Add pagination to list_expenses: add parameters 'limit' (int, default 20, max 100) and 'offset' (int, default 0). Document: 'Defaults to most recent 20 expenses. Use offset to fetch older records.' Return total_count in response so LLM knows when to paginate.
• Fix summarize 'category' parameter: change default from null to string, and document: 'category (string, optional, default=""): Filter to a specific category. Pass empty string to summarize all categories.' Then validate in code that category is either empty or in the valid categories list.