Well-structured TypeScript MCP server with 7 tools covering Seznam Webmaster API operations. All tools have clear verb-noun naming (get_*, list_, reindex_), complete input schemas with proper JSON Schema structure, and descriptions ranging 80-180 chars. Parameters are typed and described. Key strengths: consistent schema validation, proper error handling with localization, security (no credential exposure). Gaps: output schemas not documented in code, no pagination support despite tools returning potentially large result sets, no tool annotations (readOnlyHint/destructiveHint), limited parameter constraints (no enums where applicable).

• Document output schemas for all tools. Add a TypeScript interface or JSDoc @returns block showing field names, types, and descriptions. Example: 'Returns {domain: string, indexed_count: number, error_count: number, last_updated: ISO8601 string}'.
• Add tool annotations: mark read-only tools with readOnlyHint=true; mark reindex_url with destructiveHint=true. This enables agents to reason about operation safety without parsing descriptions.
• Add JSON Schema format constraints: date_from and date_to should have 'format': 'date' and pattern '^\d{4}-\d{2}-\d{2}$' to prevent LLM date format errors.
• Implement pagination for get_indexed_pages and get_index_history: add optional 'limit' (1-1000, default 100) and 'offset' parameters; return total_count and has_more in response.
• Enhance error messages with recovery hints. Instead of generic 'internal_error', return specific guidance: 'Domain not found in SEZNAM_WM_SITES. Call list_sites() to see configured domains.' This enables LLM self-correction.