Server has 3 tools with basic schemas and descriptions. All tools have input schemas with proper types (enum, number) and descriptions present. However, descriptions are minimal (10-50 chars), falling below the 50-200 char baseline for LLM-optimized descriptions. Output schemas are not documented, responses return generic text content without structured field definitions. Error handling is minimal (only divide-by-zero check in calculate tool). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Parameter descriptions lack context about valid ranges or formats. Tool composition is reasonable but descriptions do not explain when to use each tool vs. alternatives.

• Expand tool descriptions to 50-200 chars. Example: 'Performs arithmetic operations (add, subtract, multiply, divide) on two numbers. Returns the numeric result as a string. Divide by zero returns an error message.' This tells LLMs what the tool does, when to use it, and what to expect.
• Document output schemas explicitly. Add a 'returns' section to each tool description or include a schema field. Example: 'Returns {content: [{type: "text", text: string}]} where text is the menu items joined by newline for getMenu, or the numeric result for add/calculate.'
• Add readOnlyHint: true to all tool definitions in the server.tool() calls. This signals to clients that these tools are safe to call without user confirmation and can be cached/retried.
• Enhance parameter descriptions with constraints. Example: 'a: First number to add (integer or decimal, range -1e10 to 1e10)' and 'item: Menu category (burger or milkshake; case-sensitive).'
• Add error handling for invalid inputs. Return structured error responses like {content: [{type: 'text', text: 'Error: Invalid menu category. Valid options: burger, milkshake'}]} instead of silent failures or generic errors.