Four well-defined tools with complete JSON schemas, clear descriptions, and proper parameter typing. Tool names follow verb_noun convention (search_, get_, query_). Descriptions are detailed (150-250 chars) and explain WHAT, WHEN, and output semantics. All parameters have type and description. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal (no recovery guidance). No tool annotations (readOnlyHint, etc.). Composition is sound: each tool has one responsibility, names are distinct, and parameters are well-constrained.

• Document output schemas for all four tools. Example: search_events returns {events: [{id, title, canonical_summary, review_status, sources: [{book, chapter, paragraph, role}]}], total_count}. This enables LLMs to chain results into downstream calls.
• Add readOnlyHint: true to all four tool schemas to declare safety. Update inputSchema to include 'readOnlyHint': true at the tool level (MCP 2026-07-28 spec).
• Convert 'book' parameter to enum: ['shiji', 'hanshu', 'houhanshu', 'sanguozhi', 'zuozhuan', 'lunyu', 'mengzi', 'lushi-chunqiu', 'zizhitongjian']. Remove the slug list from description.
• Convert 'kind' parameter in search_events to enum: ['事件', '场景', '评价']. Remove pipe-delimited string from description.
• Add structured error responses in _dispatch_tool. Return {isError: true, message: 'Unknown tool: {name}. Available tools: search_events, get_person, query_by_place, query_by_quality'} instead of raising KeyError.