OurGroceries MCP has clear, action-verb naming and complete input schemas with proper types and descriptions. All 5 tools follow verb_noun convention (get_lists, add_item, remove_item, update_item, toggle_item). Descriptions are present and adequate (range 40 - 90 chars), though some lack context about when to use them or what they return. Parameter descriptions are consistently provided. Critical gap: NO output schemas documented, LLMs cannot predict response structure for chaining. No error handling guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles (READ_ONLY, WRITE, DESTRUCTIVE, REVERSIBLE). Schemas are well-formed with required fields and defaults, but lack examples of actual API responses.

• Document output schemas for all tools. Example for get_lists: 'Returns {lists: [{id: string, name: string, items: [{id: string, value: string, crossedOff: boolean, note: string, categoryId?: string, star: 0|1}]}]}'.
• Add tool annotations to tool definitions: readOnlyHint=true for get_lists; destructiveHint=true for remove_item; idempotentHint=true for toggle_item (safe to retry).
• Enhance error handling in CallToolRequestSchema handler: catch API errors and return structured error responses with actionable guidance (e.g., 'List not found. Call get_lists() to see available lists.').
• Expand descriptions for destructive/reversible tools: 'Remove an item from a grocery list (permanent deletion, cannot be undone)' and 'Mark an item as crossed off or uncrossed (reversible state change)'.
• Consider adding a confirm_remove_item tool or dry-run parameter to remove_item to prevent accidental deletions by agents.