Single tool with Italian descriptions and basic parameter validation. Tool has reasonable description length (195 chars) and clear naming convention (verb_noun: query_datapizza). However, output schema is completely undocumented, returns plain string with no structured format guidance. Parameter descriptions exist but lack critical details about format, constraints, and error conditions. No input validation error messages guide recovery. Error handling returns user-facing strings rather than structured error responses. Resource definitions present but peripheral to tool evaluation.

• Document output schema for query_datapizza. Return a structured object: { results: [ { title: string, content: string, source: string, relevance_score: number } ], total_count: integer, has_more: boolean, next_page: integer | null }. This allows the LLM to extract and compose results reliably.
• Add explicit parameter constraints to max_results. Change description to: 'Numero massimo di risultati (1-20, default: 5). Valori fuori intervallo verranno normalizzati a 5.'
• Convert error handling to structured format. Instead of 'return "❌ Errore: La query non può essere vuota."', return structured JSON with error code, message, and recovery hint: { error: 'empty_query', message: 'Query cannot be empty', hint: 'Provide a non-empty search term' }
• Add pagination support. Tool accepts max_results but query_documentation likely returns all results. Implement offset/limit: query_documentation(query, limit=5, offset=0) and return total_results and has_more flags so LLM can fetch additional pages.
• Convert get_system_status resource to return structured JSON instead of formatted markdown. This enables the LLM to check vector count, parse error conditions, and make decisions about re-indexing.
• Add parameter validation docstring documenting behavior for edge cases: max_results <= 0, max_results > 20, query with only whitespace, query > 500 chars.
• Include sample output in docstring showing expected structure: '{ results: [ { title: "...", relevance_score: 0.95 } ], total_count: 42 }'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry