Server exposes 7 tools with partial schema coverage. Tool names follow verb_noun convention (zhida_ask, zhida_get_session, zhihu_search_and_fetch). Descriptions present but vary in quality: zhida_ask and zhihu_search_and_fetch have detailed docstrings (200+ chars); others are brief (30-80 chars). Input schemas visible for all tools with type declarations and parameter descriptions. However, output schemas are NOT documented, LLMs cannot predict return structure. No error handling guidance, no recovery hints. No tool annotations (readOnlyHint, destructiveHint). Code shows blocking I/O (waits for stream 'End') which may cause timeout issues in agent loops. Composition is reasonable (each tool has single responsibility), but missing pagination guidance for list_sessions and search results.

• Add output schema documentation to every tool. Use JSON Schema to describe return type, fields, and field types. E.g., zhida_ask returns {question: string, turn_count: int, turns: [{role: string, content: string, ...}], ...}.
• Add readOnlyHint: true to all tool schemas (via tool annotations in MCP 2026-07-28). This signals to agents that these tools are safe to retry and have no side effects.
• Wrap error responses with actionable recovery hints. E.g., 'Invalid URL: must be zhihu.com/question/... or zhihu.com/p/... . Try zhihu_search_and_fetch to find valid URLs.' instead of bare ValueError.
• Add timeout parameter to zhida_ask (default 30s) and document it. Return timeout error with retry guidance if stream does not close within limit.
• Enumerate valid values for constrained parameters. Add 'chat_model' enum: [CM_DEEP_SEEK_R1, CM_GPT_4, ...] and 'kinds' enum: [answer, article, question] to tool schemas.