mail-use provides 20 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention (email_list, email_send, email_delete). Descriptions are present and substantive (avg ~180 chars), exceeding the 10-char minimum. However, several critical gaps reduce the score: (1) Parameter descriptions are inconsistent, many reuse boilerplate text ('Account id (e.g. 'leeguooooo_gmail')...') across tools, reducing clarity; (2) Output schemas are not documented, callers cannot predict response structure; (3) Error handling lacks recovery guidance; (4) Destructive operations (email_delete, cleanup_apply) lack dry_run confirmation patterns in all cases; (5) Some parameters like 'ids' in email_show lack type specificity (array of what?). The server does accept human-friendly identifiers (email addresses, account names) alongside IDs, which is good. Zod schema validation is present but not exposed to LLMs.

• Document output schemas for all tools. Specify response fields, types, and structure. Example: email_list returns {emails: [{id, from, subject, date, unread}], total_count, folder_unread}.
• Add explicit dry_run confirmation pattern to email_delete and cleanup_apply descriptions: 'Pass dry_run=true to preview changes before applying. Always preview destructive operations.'
• Shorten and differentiate account_id parameter descriptions. Instead of repeating the same text, use: 'Account identifier (e.g., leeguooooo_gmail, [email protected])' and reference account_list for discovery.
• Specify array element types in schemas. Change 'ids': {type: 'array'} to 'ids': {type: 'array', items: {type: 'string'}, description: 'Email UIDs or gids (e.g., ["123", "456"])'}.
• Add error classification to tool descriptions. Example: 'Returns 404 if account not found (user-fixable: call account_list to verify). Returns 500 on IMAP timeout (retryable).' This guides LLM recovery logic.