The server provides three well-structured tools with proper Zod schemas and reasonable descriptions. Naming follows verb_noun convention (web_search, web_scrape, screenshot). Descriptions are concise (60-80 chars) and mention pricing/zero-logging, which is helpful context. However, descriptions are at the lower end of the 194-char baseline and lack critical metadata about when to use each tool vs alternatives. Parameter descriptions are present and constrained with min/max ranges, but several lack detail about expected formats. Error handling returns basic error messages but does not provide recovery guidance per pattern:recovery-guide. The tools form a coherent composition for web intelligence gathering, but lack downstream chaining information (e.g., what IDs are returned that other tools might need). No output schemas are documented.

• Expand tool descriptions to 150 - 250 chars. Include: (1) What it does, (2) When to use it (vs web_search for discovery, vs web_scrape for deep content, vs screenshot for visual verification), (3) What it returns (e.g., 'Returns list of results with title, URL, and snippet'), (4) Cost and rate-limit info.
• Document output schemas explicitly. For web_search, specify: { results: Array<{ title, url, description, position }> }. For web_scrape: { title, content, timestamp }. For screenshot: { data (base64 string), mimeType }.
• Add recovery guidance to error handling. Example: catch block for web_scrape URL errors should return 'Invalid URL format, ensure it starts with http:// or https://. Got: ${url}'. For timeout errors: 'Page took too long to load (>30s). Try increasing the timeout parameter or checking if the URL is responsive.'
• Expand parameter descriptions to include format hints and constraints. Example: query: 'Search query string (up to 500 chars, supports AND/OR/NOT operators)'. url: 'Valid HTTP/HTTPS URL. Pages >50MB will be truncated; JavaScript content may not render.'
• Add a cost_estimate or dry_run parameter to screenshot and web_scrape tools. Allow agents to preview the cost before executing: { dryRun: true } returns estimated cost without charging. This prevents runaway agent loops from incurring unexpected expenses.
• Add pagination to web_search results. Declare: count param (1 - 20 results per page, default 10), offset param (0-based), and return { results: [...], total_count: N, has_more: bool }. This follows pattern:paginated-result and allows agents to fetch all results iteratively.