Five tools with clear, domain-specific names and solid descriptions (avg 180 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are not formally documented, responses are JSON strings without declared field structures. Parameter descriptions lack format constraints (e.g., ticker format, video_id range). Error handling is minimal; tools return generic JSON errors without recovery guidance. No tool annotations (readOnlyHint, etc.). Composition is sound: tools chain logically (search → lookup → load). Naming follows verb_noun pattern well (query_, get_, lookup_, search_, load_). Missing: pagination guidance for search_transcripts (max_results=25 is capped but no cursor/offset), output schema documentation, and actionable error messages.

• Add formal output schema documentation for each tool. E.g., query_trading_rules returns {count: integer, rules: [{category: string, rule_text: string, frequency: integer, source_videos: [integer]}]}. Declare all fields, types, and whether they are always present.
• Convert parameter descriptions to include format constraints. E.g., 'ticker: Stock ticker symbol (uppercase, 1-5 characters, e.g., ROKU, TSLA)' and add enum constraint in schema for category parameter.
• Enhance error messages with recovery guidance. E.g., 'Knowledge base not yet built. Run the extraction pipeline (see README) or check that knowledge/ directory contains transcripts.json, rules.json, setups.json, criteria.json, ticker_index.json, topic_index.json.'
• Add tool annotations via FastMCP. Mark all tools with readOnlyHint=true since they only query data. This signals to LLMs that tools are safe to retry and do not modify state.
• Implement cursor-based pagination for search_transcripts. Return {count, results, next_cursor} and accept optional cursor parameter. This enables LLMs to fetch large result sets incrementally without re-querying.