Thomisidae provides three web-scraping tools with explicit schemas and generally clear naming. However, several definition quality gaps prevent a higher score. All three tools have descriptions (though some are generic), and all have JSON Schema input definitions. Parameter descriptions are present but inconsistent in depth, some parameters like 'mode' in fetch_url_text lack actionable guidance on when to use each enum value. Output schemas are undocumented (not visible in the code provided), which violates pattern:tool-output-schema. Error handling is basic: the code shows validation errors (e.g., 'Invalid url', 'Blocked by robots.txt') but lacks recovery guidance or actionable next steps for agents. The tool names are clear and verb-first (fetch_*, google_search), meeting naming baseline standards. However, parameter naming has minor inconsistencies: 'raw' could be clearer as 'return_raw_content', and the 'respect_robots' override parameter is not self-documenting.

• Document the return schema for each tool. Example for fetch_url_text: 'Returns a ToolResult with a single text field containing the extracted content, prefixed with URL and optional title.'
• Enhance parameter descriptions with use-case guidance. For 'mode' in fetch_url_text, specify: 'auto: intelligently chooses extraction method. best_blocks: extracts semantically coherent text blocks. readability: Mozilla readability algorithm (requires readability feature). raw: returns unprocessed HTML.'
• Add recovery hints to error messages. When robots.txt blocks access, return: 'Access to {url} blocked by robots.txt. If this is a trusted internal resource, retry with respect_robots=false.'
• Clarify fetch_page_links output. Does it return absolute URLs or relative? How are duplicates handled? What does 'text' format return, plain link text per line, or something else?
• Change 'respect_robots' from nullable boolean to required boolean with explicit default. Remove the null union type.
• Document max_length and start_index chunking behavior. Does truncation break mid-word? How are overlaps handled if start_index + max_length > content length?