Strong tool naming (all verb_noun pattern) and comprehensive parameter schemas with types and descriptions. All 10 tools have clear, actionable descriptions (avg 150 chars). Pagination support via cursor/limit/returnPage is well-designed. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is absent from visible code. No tool annotations (readOnlyHint, idempotentHint). Parameter descriptions could be more prescriptive about constraints (e.g., 'limit capped globally' is vague). Missing guidance on when to use mail_search vs mail_search_advanced vs mail_query_by_folder, three similar tools risk LLM confusion.

• Document output schemas for all tools. Example: mail_list_messages should return {items: [{uid, from, subject, date, snippet?}], nextCursor?, total?}. Include field types and whether fields are always present.
• Consolidate search tools: keep mail_search_advanced as the canonical search, remove mail_search and mail_query_by_folder, or clearly document when each is preferred (e.g., 'Use mail_query_by_folder for simple free-text, mail_search_advanced for structured filters').
• Replace vague limit descriptions with explicit caps: 'Max results (default 50, global cap 500)' instead of 'capped globally'. State the actual number.
• Add tool annotations: set readOnlyHint=true and idempotentHint=true on all tools so agents know they are safe to retry and do not modify state.
• Add error recovery guidance to descriptions: 'If mailbox not found, call mail_list_folders first to see available folders.' This guides agents on recovery paths.