The Gastos server has 2 tools with mixed quality. Both tools have descriptions and input schemas, but several critical gaps exist: (1) tool names lack clear action verbs that distinguish function from data retrieval; (2) parameter descriptions in the schema are present but vary in quality; (3) no error handling guidance is provided, errors return raw exception strings rather than actionable recovery hints; (4) output schemas are undocumented, neither tool's response structure is formally specified; (5) parameter ranges and constraints are absent (e.g., 'dias' has no bounds); (6) the tools do not follow clear composition patterns (obtaining recent expenses should not require a separate tool when the data could be enriched in responses); (7) security issues: CSV file path is hardcoded with no validation against path traversal, and the agregar_gasto tool performs unvalidated file I/O. These are significant gaps for a production-grade MCP server.

• Add formal output schema documentation for both tools. Example for obtener_gastos_recientes: 'Returns: list[dict] where each dict has keys {fecha: str (YYYY-MM-DD), categoria: str, cantidad: float, metodo_pago: str}'. For agregar_gasto: 'Returns: dict with keys {success: bool, message: str, expense_id?: str}'.
• Add numeric constraints to parameter descriptions and JSON Schema. Example: 'dias (integer, minimum=1, maximum=365, default=5): Number of past days to query.'
• Define an enum for 'categoria' in agregar_gasto. Example: 'categoria (string, enum=["Food", "Transport", "Entertainment", "Utilities", "Other"]): Expense category.'
• Rename tools to English with clear action verbs: 'agregar_gasto' → 'add_expense', 'obtener_gastos_recientes' → 'get_recent_expenses'.
• Replace raw exception strings with actionable error messages. Example: Instead of 'Error al agregar gasto: [Errno 2] No such file or directory', return {success: false, message: 'Failed to add expense: data/ directory does not exist. Create it or contact admin.', recoverable: true}.
• Add a dry_run parameter to agregar_gasto (boolean, default=false). When true, validate the input and return a preview without writing to the CSV.
• Add limit and offset parameters to obtener_gastos_recientes. Example: limit (integer, 1-100, default=20), offset (integer, default=0). Return also a total_count so LLMs know whether more results exist.
• Validate the 'fecha' parameter format before processing. Return: 'Invalid date format: expected YYYY-MM-DD, got {fecha}' if parsing fails.