Server has 10 tools with complete schemas and descriptions in Italian. Naming follows verb_noun pattern (lista_, leggi_, cerca_, esegui_, salva_). Descriptions are present (avg ~100 chars) but lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Parameters are typed and described. Critical gap: no error handling guidance (e.g., what to do if PDF is corrupted, preset not found). Batch/composition tools (esegui_comandi, esegui_preset) are well-designed but lack idempotency documentation. Output schemas are implicit in code, not formally documented. Security: tools accept file paths (nome_file), path traversal mitigated by store.percorso_sicuro() but not documented in tool descriptions.

• Add explicit output schema documentation to each tool. Example: 'Returns: {"cartella": string, "esiste": boolean, "cedolini_pdf": integer, "presenze_csv": integer, "suggerimento": string|null}'
• Enhance error descriptions with recovery guidance. Example for leggi_cedolino: 'If PDF is corrupted or unreadable, call lista_cedolini() to verify the file exists and is accessible.'
• Document the path traversal mitigation in leggi_cedolino description: 'File paths are validated against directory traversal attacks; only files within INAZ_DATA_DIR are accessible.'
• Add idempotency guarantees to esegui_comandi: 'Each command is executed independently; if a command fails, previously executed commands retain their effects. Retrying the batch is safe but may re-execute successful commands.'
• Provide English translations of all tool descriptions and parameter names to improve LLM discoverability and reduce language-based misinterpretation.