ZeroCrawl has 7 well-named, action-verb tools with complete input schemas and descriptions. All tools follow verb_noun naming (scrape_page, crawl_domain, extract_metadata, search_crawl, parse_sitemap, take_screenshot, batch_scrape). Descriptions are present and contextual (avg ~80 chars), exceeding the 10-char minimum. However, parameter descriptions are sparse or missing entirely, most params lack detail on format, constraints, or expected values. Output schemas are not documented in the source, forcing LLMs to infer result structure. Error handling is minimal; no recovery guidance or categorization. No pagination support despite tools returning lists. Security is adequate (no exposed secrets in params), but audit logging is absent.

• Document output schemas for all 7 tools. E.g., scrape_page should return {url: string, markdown: string, title?: string, error?: string}. This enables LLMs to chain results into downstream tools.
• Expand parameter descriptions to 50 - 100 chars with format/constraint details. E.g., 'url: Full HTTP(S) URL to scrape (max 2048 chars). Supports redirects. Returns error if unreachable after 3 retries.'
• Add error categorization and recovery guidance. E.g., 'If scrape fails with timeout, try take_screenshot instead. If URL is blocked, use search_crawl to find cached/alternative sources.'
• Implement pagination for search_crawl and batch_scrape. Add limit (default 20, max 100) and offset/cursor parameters. Return total_count and next_cursor in responses.
• Document batch_scrape's 5-URL limit in the tool description: 'Accepts up to 5 URLs per call. For larger batches, make multiple calls or use crawl_domain for recursive discovery.'