Server exposes 9 tools with reasonable naming (verb_noun pattern: search_holes, get_hole, browse_by_tag, list_hot_tags, browse_division, list_divisions, search_courses, get_course_reviews, check_connection). All tools have descriptions in Chinese with context about when to use them. Input schemas are present with types and defaults. However, descriptions lack English translations (barrier for non-Chinese-speaking agents), parameter descriptions are minimal (e.g., 'Tree hole ID' vs actionable guidance), and output schemas are not formally documented. Error handling returns user-friendly messages but lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint). Composition is sound, tools chain logically (search_holes → get_hole, search_courses → get_course_reviews). Pagination is implemented (limit params capped at 50). No security issues detected (no secrets in params, read-only operations).

• Add English translations to all tool descriptions and parameter docs. Use bilingual format: '在复旦树洞全站搜索 (Search Fudan Tree Hole forum)' so agents in any language can understand intent.
• Document output schemas formally. Define return type as JSON object with fields: hole_id (int), time_created (string), reply_count (int), tags (array), content_preview (string). Update _fmt_hole and _fmt_floor to return structured dicts, not formatted strings.
• Add tool annotations to all tools: @mcp.tool(readOnlyHint=True) for all 9 tools. This signals to agents that these are safe to call without confirmation.
• Enhance parameter descriptions with constraints inline. E.g., 'limit: Maximum number of results (1 - 50, default 15)' and 'keyword: Search term (1 - 100 chars, supports Chinese and English)'.
• Improve error messages with recovery guidance. E.g., 'Captcha triggered. Wait 5 minutes before retrying. If persistent, call check_connection() to verify credentials.' Categorize errors as retryable vs user-fixable.