Three tools with complete schemas and descriptions, but descriptions are minimal (10-20 chars baseline). All tools are READ_ONLY and properly parameterized with types. search_customers and get_order_by_id have good input validation logic (limit clamping, SQL parameterization). get_weather lacks error recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. Output schemas are not documented, LLMs cannot predict response structure. Parameter descriptions are present but terse (e.g., 'Nombre o parte del nombre del cliente a buscar' is 47 chars, below the 72-char baseline). No pagination support on search_customers despite potentially returning multiple results.

• Add output schema documentation to each tool. E.g., search_customers returns list[{id: int, name: str, email: str}]. get_order_by_id returns {id: int, customer_id: int, customer: str, total: float, status: str} or null.
• Apply readOnlyHint=true to all three tools in FastMCP decorator: @mcp.tool(readOnlyHint=True). Signals to agents these are safe to call without side effects.
• Expand search_customers description to include pagination guidance: 'Searches customers by partial name match (case-insensitive). Returns up to `limit` results (default 10, max 50). Use limit parameter to control result count.'
• Enhance get_weather error messages with recovery hints. Instead of {'error': 'No encontré la ciudad'}, return {'error': 'City not found. Try a larger region (e.g., "Buenos Aires" instead of "B.A.") or check spelling.'}
• Document parameter constraints in descriptions. E.g., limit: 'Maximum results to return (1 - 50, default 10). Larger limits may slow queries.' order_id: 'Numeric order ID. Returns null if order does not exist.'