Two tools with minimal documentation. Both have basic descriptions (50-60 chars) and input schemas present but sparse. 'extract' and 'search' lack parameter descriptions entirely, the schema shows only type and top-level description, no per-parameter guidance. No output schema documented. No error handling guidance. Naming is verb-based but generic ('extract', 'search'). No evidence of idempotency, pagination, or composition patterns. Tools appear inferred from API wrapper code rather than explicitly registered with full MCP metadata.

• Add per-parameter descriptions to both tools. For 'extract': describe 'url' format (must start with http:// or https://), and 'schema' as a JSON Schema object defining fields to extract. For 'search': describe 'query' as plain-language ICP description, and 'num_results' as integer 1-50 (default 10).
• Document output schemas. 'extract' should return {data: {[field]: value}, error?: string}. 'search' should return {results: [{name, website, one_liner, source}], total_count, next_cursor?}.
• Add error handling guidance. E.g., 'If extraction fails, the URL may be unreachable or the schema may not match the page structure. Try a simpler schema or verify the URL is public.'
• Rename tools for clarity: 'extract_website_data' and 'search_listing_pages' better convey intent and domain.
• Add pagination support to 'search': accept 'page' or 'offset' parameter, return 'total_count' and 'next_cursor', and document max results per page (e.g., 50).