Server has 4 tools with basic schemas but significant quality gaps. Tool descriptions are present but minimal (10-80 chars, below the 50-200 char baseline for LLM optimization). Parameter descriptions exist but are sparse. No output schemas documented. Error handling is absent, no recovery guidance, no categorization, no actionable error messages. Tools lack idempotency hints and composition support. The query_data tool has a complex filters parameter accepting multiple types (object|array|string|null) without clear guidance on format or usage.

• Expand tool descriptions to 50-200 characters. Example: 'get_data_sources: Retrieve all available data sources with their IDs and metadata. Call this first to discover which sources you can query.' (English; current Russian descriptions should be translated or localized consistently.)
• Document output schemas for all tools. Example for get_data_sources: 'Returns {"sources": [{"id": string, "name": string, "type": string}]}'. This enables LLMs to chain tools (e.g., extract source_id from response, pass to get_dimensions).
• Clarify query_data filters parameter. Add enum or pattern constraint: 'filters: object (e.g., {"region": "US", "year": 2024}) or array of objects for OR logic, or JSON string. Omit for no filtering.'
• Add error handling with recovery hints. Wrap tool calls in try-catch; return structured errors: '{"error": "source_not_found", "message": "Source 'invalid_id' not found.", "recovery": "Call get_data_sources() to list available sources."}'
• Add tool annotations to the FastMCP tool registration. Mark all tools with readOnlyHint=true and idempotentHint=true so agents know they are safe to retry without side effects.