The server defines 3 tools with partial schemas and descriptions. Tool names follow verb_noun convention (search_, get_), which is good. However, descriptions are generic (10-60 chars), parameter descriptions are minimal or missing entirely, and output schemas are completely undocumented. The server lacks error handling guidance, has no input validation documentation, and provides no recovery paths. Definitions are visible in src/index.ts and properly registered via ListToolsRequestSchema, but fall well below production baselines (A+ tools average 194 chars for descriptions; these average ~45 chars). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present, though all three tools are READ_ONLY. The schema quality is inconsistent: search_with_fallback_targetted_websites has good parameter constraints (enums, min/max), but get_full_content and get_latest_defi_news lack output documentation entirely.

• Document output schemas for all three tools. Specify what fields are returned (e.g., 'Returns an array of {title: string, url: string, snippet: string, published_date: string}'). Include example structures in tool descriptions.
• Expand descriptions to 100-200 chars following pattern:tool-description. For get_full_content: 'Fetch and parse the full HTML content from a URL. Returns plaintext or markdown. Use this to read article bodies after search_with_fallback_targetted_websites returns URLs. Useful for extracting detailed DeFi news content.'
• Add readOnlyHint annotation to all three tools. In the Tool object, add 'annotations: { "io.modelcontextprotocol/hints": { "readOnly": true } }' to signal safety.
• Enhance parameter descriptions. For 'search_depth' enum values, add: 'basic=surface-level DeFi website search; advanced=multi-layer search with fallback to Tavily general search.' For 'limit' in get_latest_defi_news, clarify: 'Maximum number of news articles to return (cached, updated every 24h).'
• Add error handling examples. In CallToolRequestSchema error handler, return contextual guidance: e.g., for URL extraction errors: 'Failed to extract content from {url}. Try search_with_fallback_targetted_websites to find alternative sources.'; for API timeouts: 'Tavily search timed out. Retry with search_depth=basic or reduce max_results.'
• Document result limits and sorting. For get_latest_defi_news: 'Results are cached from the last 24h, sorted by publish date (newest first). Limit capped at 50 to preserve context window. For full archive, use search_with_fallback_targetted_websites.'