The server has 2 well-named tools with solid descriptions and documented parameter schemas using Zod validation. However, output schemas are not formally documented, error handling lacks recovery guidance, and composition could be improved. Tool names follow verb_noun convention (list_, search_) appropriately. Descriptions are adequate (100-150 chars) but lack specificity about when to use each tool vs. the other. Parameters are typed and described, but some constraints (like date format ISO 8601) could be more explicit. No tool annotations present.

• Document output schemas explicitly. For list_meetings, specify: 'Returns {total_found: number, showing: number, meetings: Array<{title: string, date: string (ISO 8601), url: string, attendees: string[], summary: string, action_items: string[], transcript?: string}>, has_more: boolean}.'
• Add tool annotation in server registration: {name: 'list_meetings', ..., readOnlyHint: true, idempotentHint: true}. Same for search_meetings.
• Enhance error messages with recovery guidance. Example: 'Invalid date format. Use ISO 8601, e.g., "2024-01-15T10:30:00Z". Or call list_meetings() without date filters to see all meetings.'
• Clarify when to use each tool in descriptions. Update search_meetings description to: 'Search for specific keywords across meeting titles, summaries, or action items (searches last 30 days). Use list_meetings() to filter by date, attendees, or team instead.'
• Add pagination to search_meetings. If results exceed 50, add offset and limit parameters, or return a next_cursor. Cap default results at 50 to prevent context window exhaustion.
• Tighten parameter descriptions with format examples: 'created_after: ISO 8601 date-time string (e.g., "2024-01-15" or "2024-01-15T10:30:00Z"). Meetings created after this date are included.'
• Reduce response verbosity. Strip internal fields and return only: title, date, attendees, recorded_by, summary, action_items, url, and optional transcript. Remove metadata that clutters the LLM context.