Server has 10 tools with basic schemas and descriptions in Chinese. All tools have input schemas with type definitions and descriptions present. However, descriptions are minimal (10-50 chars), lack context on WHEN to use tools, and do not explain prerequisites or error recovery. Parameter descriptions are present but terse. Output schemas are not documented, LLMs cannot predict return structure. No error handling guidance. Tool names follow verb_noun pattern (list_*, read_*, write_*, update_*, delete_*, add_*, reply_*) which is good, but descriptions lack the depth needed for reliable LLM selection. Security-sensitive tools (write_doc, update_doc, delete_doc) lack confirmation/dry-run patterns. No pagination support on list_* tools despite potential for large result sets.

• Expand all tool descriptions to 100-200 characters. Include: (1) what the tool does, (2) when to use it vs similar tools, (3) any prerequisites (e.g., 'Requires write permission'), (4) what it returns. Example: 'Create a new document in your role space. Use this after list_roles() to confirm your role_id. Returns document ID for use in update_doc or delete_doc. Fails if title already exists in your role.'
• Document output schemas for all tools. Add a 'Returns' section to each docstring listing field names, types, and meanings. Example for list_docs: 'Returns list of {id: int, title: str, role_id: int, version: int, updated_at: ISO8601 string}'.
• Add pagination to list_* tools: accept optional 'limit' (default 20, max 100) and 'offset' (default 0) parameters. Return total_count and has_more fields. This prevents context explosion on large datasets.
• Implement a confirm_delete pattern for delete_doc: add an optional 'confirm=false' parameter. If false, return a preview of what will be deleted. If true, delete. This prevents accidental destruction.
• Enhance error messages with recovery guidance. Instead of 'PermissionError("未认证")', return 'Authentication failed. Ensure your token is valid and not expired. If you just created the token, wait 30 seconds for propagation.' This guides LLM retry logic.