Five tools with complete schemas and descriptions in Russian. Naming follows verb_noun pattern (screen_*, list_*). Descriptions are domain-specific but lack LLM-optimized guidance on when/why to use each tool. Parameters are well-typed with constraints (minLength, maxLength, nullable). No output schemas documented. Error handling returns structured dicts with disclaimers but lacks recovery guidance. Missing tool annotations (readOnlyHint present in metadata but not in schema). No batch operations or pagination support despite potential for large result sets.

• Document output schemas for all tools. Example for screen_person: {matches: [{name, confidence: 'exact'|'partial'|'possible_namesake', source, url}], disclaimer}. Enables LLM to extract and chain results.
• Enhance descriptions with 'when to use' guidance. E.g., screen_person: 'Use when you have a full name but no INN. For faster exact matching with INN, use screen_inn instead.' Helps LLM choose the right tool.
• Add recovery hints to error responses. E.g., backend_error → 'The screening service is temporarily unavailable. Retry in 30 seconds or contact [email protected].' Guides agent retry logic.
• Verify tool annotations (readOnlyHint) are serialized in FastMCP's JSON Schema output. If not, add explicit 'readOnly: true' to schema or document in description.
• Add optional limit and offset parameters to list_watchlists. Cap default to 50 items. Document in description: 'Returns up to 50 watchlists per page. Use offset for pagination.'

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

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