Server has 20 tools with mostly complete schemas and descriptions in Russian. Naming follows verb_noun pattern (find_, create_, list_, etc.). Descriptions are substantive (100-300 chars typical) and explain WHEN to use tools (e.g., 'Вызывать ДО create_bug_report'). However, several critical gaps: (1) Parameter descriptions lack format/constraint details (e.g., 'threshold' has no range guidance; 'impact' enum values documented only in description, not schema); (2) Output schemas are not formally documented, responses inferred from code; (3) Error handling is minimal, no recovery guidance or categorization; (4) Some parameters accept free-form strings where enums would be safer (e.g., 'status', 'category', 'reply_kind'). Tool composition is strong, each tool has one responsibility and outputs chain well (e.g., find_duplicate_issue → link_ticket_to_issue or create_bug_report). Descriptions are LLM-optimized and include dependency hints.

• Add JSON Schema enum constraints for all categorical parameters: status, impact, urgency, reply_kind, severity, category. Move from description-only to formal schema enums.
• Document output schemas for all 20 tools. Create a schema definition for each response (e.g., {duplicates_found: integer, candidates: array, recommendation: string}) and include in tool metadata.
• Add numeric bounds to all numeric parameters: threshold 0.0 - 1.0, top_k 1 - 50, limit 1 - 100, offset ≥0, eta_minutes 1 - 1440, confidence 0.0 - 1.0.
• Enhance error responses with recovery guidance. Replace generic errors with actionable messages: 'Ticket not found. Call list_tickets(status="Новая") to see available tickets.' or 'Invalid impact value. Must be: single_user | team | org_wide'.
• Add total_count and next_cursor to paginated responses (list_tickets, kb_search, list_issues) so agents know when to stop iterating and can resume efficiently.