Server has 4 tools with Russian-language descriptions and basic parameter schemas. All tools are READ_ONLY and stateless. However, critical gaps exist: (1) All descriptions are in Russian, making them inaccessible to most LLM agents trained on English; (2) Parameter descriptions lack detail on format/constraints (e.g., 'dataset' param has no guidance on valid format or how to discover valid IDs); (3) No output schemas documented, LLMs cannot predict response structure; (4) Error handling is minimal (e.g., dataset_versions returns a string message for not-found, but no structured error); (5) No parameter validation hints (e.g., salary_history accepts 'subject' as free-form string with no enum or pattern). Tools are simple and well-intentioned but lack production-grade polish.

• Translate all tool descriptions and parameter descriptions to English. Use clear, concise English (50 - 200 chars per description) optimized for LLM parsing. Example: 'List all available ФГИС ЦС datasets with version counts and date ranges. Call this first to discover dataset IDs for other tools.'
• Document output schemas for each tool. Define structured JSON responses with typed fields. Example for list_datasets: {datasets: [{id: string, title: string, versions: int, first_date: string, last_date: string, schema_count: int, size_mb: float}], source: string}
• Add parameter constraints and discovery hints. For 'dataset' param: 'Dataset ID (format: XXXXXXXX-XXXXX, e.g., 7707082071-ksrms). Call list_datasets() to discover valid IDs.' For 'subject': 'Region name or substring (e.g., Иркутская, Москва). Partial matching supported.'
• Implement structured error responses. Return JSON with {error: string, code: string, recovery: string} instead of plain text. Example: {error: 'Dataset not found', code: 'DATASET_NOT_FOUND', recovery: 'Call list_datasets() to see available datasets.'}
• Add per-tool output examples in descriptions. Example for salary_history: 'Returns salary history for matching regions with dates, rates, and growth percentages. If multiple regions match, returns all with separate sections.'