Server provides 4 well-named, read-only discovery tools with complete input schemas and clear descriptions. All tools follow verb-noun naming convention (list_, get_, find_, chat_with_). Descriptions are adequate (90-180 chars) and explain WHEN to use each tool. Input schemas are properly typed with enums where applicable. However, output schemas are completely undocumented, the code returns JSON.stringify() with no structured schema definition visible to clients. Error handling is absent: no guidance on what happens if an ID is invalid, a country name doesn't match, or the service is unavailable. The tools are well-scoped and compositional (list → get is a natural chain), but lack the production-grade output documentation and error recovery patterns expected in a 70+ score.

• Document output schema for each tool. Example for list_ai_services: 'Returns array of objects with fields: id (string, service identifier), name (string, display name), country (string), provider ("voxie" | "third-party"), description (string), webchat (object with url property). Limit: 20 results.'
• Add explicit error handling. For get_ai_service, return: '{"error": "Service not found", "suggestion": "Try list_ai_services() to see available IDs."}' when service is not found.
• Enhance parameter descriptions with format constraints. For 'country', change to: 'Country name or partial match. Case-insensitive, minimum 2 characters (e.g., "US", "Canada", "united states").'
• Add discovery hints to descriptions. For get_ai_service, append: 'Call list_ai_services() first to discover valid service IDs.'
• Document pagination behavior if large result sets are possible. For list_ai_services, specify: 'Returns up to 20 services. Use provider filter to narrow results if more are available.'
• Validate inputs before processing. Check that country name has at least 2 characters and service ID matches known format to provide early feedback.