Server has 21 well-named tools with consistent verb_noun patterns (mail_list_*, mail_create_*, mail_delete_*, etc.). All tools have descriptions and input schemas with Zod validation. However, descriptions are generic and lack LLM-optimization guidance (e.g., 'List emails in a folder' vs 'Retrieve emails from a specific folder with pagination; use limit/offset for large mailboxes'). Output schemas are not documented, responses are JSON-stringified but structure is opaque to the LLM. Parameters lack constraints (enums, ranges, format hints). Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Security: MAIL_TOKEN is injected via environment, which is correct, but no audit logging or permission gates are visible. Composition is sound, each tool has one responsibility, and IDs chain correctly (mailbox_uuid, folder_id, message_id flow through calls).

• Document output schemas for all tools. E.g., mail_list_mailboxes should return {mailboxes: [{uuid: string, name: string, email: string}], total: number}. Include field types and descriptions so LLM can chain calls.
• Add confirmation/dry-run pattern to destructive tools. E.g., mail_delete_emails should accept a 'confirm' boolean parameter; return {status: 'pending_confirmation', message: 'Ready to delete 5 emails. Call again with confirm=true to proceed.'}
• Enhance parameter descriptions with constraints and guidance. E.g., 'limit: Maximum emails to return (1-100, default 50). Use lower values for faster responses on large mailboxes.' Include WHEN to use each parameter.
• Implement error handling with recovery guidance. E.g., if folder not found, return {error: 'Folder not found', code: 'FOLDER_NOT_FOUND', suggestion: 'Call mail_list_folders to see available folders.'}
• Add input validation and constraints. E.g., limit: z.number().min(1).max(100).default(50), since: z.string().regex(/^\d{4}-\d{2}-\d{2}$/).optional(). Return actionable errors like 'Invalid date format: expected YYYY-MM-DD, got 2024/01/15'.