Both tools have minimal descriptions (under 100 chars) and lack comprehensive parameter documentation. Input schemas are present but sparse: google_search_scraper has 2 params with basic types; universal_scraper has 3 params with basic types. No output schemas documented. No error handling guidance. Tool names are verb-noun but descriptions are too terse to guide LLM selection. Parameters lack constraints (enums, ranges, format hints). No indication of idempotency, side effects, or retry safety. Schemas lack the depth needed for production agent composition.

• Expand tool descriptions to 100 - 200 characters. For google_search_scraper: 'Search Google for a query and optionally parse results into structured data (title, URL, snippet). Use parse=true to get machine-readable output; parse=false returns raw HTML. Returns up to 10 results per page.' For universal_scraper: 'Scrape a URL and return content in the specified format (markdown, HTML, or plaintext). Set render=html to execute JavaScript before scraping. Returns page content, metadata, and status. Useful for extracting article text, product details, or structured data from web pages.'
• Document output schemas for both tools. Specify: google_search_scraper returns {results: [{title, url, snippet, parsed_content?}], total_results, page}. universal_scraper returns {content, format, status, metadata: {title?, author?, publish_date?}, error?}. Include field types and whether fields are optional.
• Convert parameter examples to enum constraints. For output_format, use enum: ['md', 'html', 'plaintext', 'json']. For render, use enum: ['none', 'html', 'javascript']. Add descriptions: 'output_format: Output format for scraped content. Use md for markdown (best for LLM readability), html for raw HTML, plaintext for text-only, json for structured extraction.'
• Add error handling guidance to descriptions. E.g., 'If no results found, try broadening the query or check internet connectivity. If universal_scraper returns empty content, try render=html to load JavaScript-heavy pages. If rendering fails, the page may require authentication or be blocked.'