Strong schema definitions with proper JSON Schema types and enums across all 12 tools. Tool names follow verb_noun convention (get_*, search_*). Descriptions are present and substantive (avg ~180 chars), exceeding the 10-char minimum. Pre-call validators enforce format constraints (OGRN 13 digits, INN 10/12 digits, BIC 9 digits). However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions are detailed but some lack explicit format guidance (e.g., date_from/date_to state 'YYYY-MM-DD' but not whether partial dates are accepted). Error handling is minimal, no recovery guidance or actionable error messages visible in tool definitions.

• Document output schemas for all tools. Example: get_company returns {company_id, name, inn, ogrn, directors: [{name, role, inn}], ...}. This enables LLMs to chain tools and extract required fields.
• Add error recovery hints to tool descriptions. E.g., 'If OGRN is invalid, call search() with the company name to find the correct OGRN.' This guides LLM self-correction.
• Specify pagination behavior explicitly: 'Returns up to 100 results per page. Include total_count in response so LLM knows if more pages exist.'
• Tighten date format descriptions: 'date_from must be ISO 8601 (YYYY-MM-DD). Partial dates like 2024-01 are not accepted.' This prevents LLM format errors.
• Add tool annotations to the MCP schema: mark all tools with readOnlyHint=true and idempotentHint=true so LLMs know they are safe to retry without side effects.