Server has 7 tools with basic descriptions and input schemas, but significant gaps in parameter documentation, output schema clarity, and error handling. Tool naming is mostly verb-driven (search, read, index, ask_gpt, pipe_chat_to_gpt) but lacks consistency in parameter naming conventions. Descriptions range from adequate (search: 180 chars) to vague (index: lacks detail on what 'model/schema changes' means). Most parameters lack type constraints (enums, ranges). Output schemas are not documented. Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. Security concern: ask_gpt and pipe_chat_to_gpt expose file attachment and transcript reading without clear sandboxing or permission gates.

• Document output schemas for all tools. E.g., search should return {results: [{title, snippet, path, score}], total_count, next_cursor}. read should return {content, path, size_chars, truncated}.
• Add parameter constraints: search/search_wiki limit should be 'integer, 1-100, default 8'. read/read_wiki_page path should be 'string, relative path format: folder/file.md, no ../ traversal'.
• Add error handling guidance to descriptions. E.g., search: 'Returns empty results if no matches found. If expected results missing, try broader query or call index() to rebuild.' ask_gpt: 'Returns error if OpenAI API key invalid or quota exceeded, check GPT_MCP_KEY_FILE and OPENAI_API_KEY env vars.'
• Implement path traversal validation in ask_gpt/pipe_chat_to_gpt file attachment logic. Reject paths containing '..' or absolute paths. Document sandbox scope in tool descriptions.
• Add tool annotations (readOnlyHint, destructiveHint) to clarify side effects. search, read, search_wiki, read_wiki_page should be marked readOnly. index should be marked destructive (rebuilds index).