The server provides 4 well-defined search and scraping tools with proper Zod schemas and descriptions. Tool naming is clear and action-oriented (one_search, one_scrape, one_map, one_extract). Descriptions are adequate but could be more LLM-optimized. All tools have input schemas visible in src/schemas.js (imported as SearchSchema, MapSchema, ScrapeToolInputSchema, ExtractSchema). Parameter schemas are well-structured with types, descriptions, and constraints (enums, optional flags). However, output schemas are not documented in the provided source, the tool definitions only show input schemas. Error handling uses a wrapper pattern (createToolHandler) with structured error responses (isError flag + text content), but lacks recovery guidance or categorization. Security is reasonable (read-only tools, no credential exposure), but tool descriptions could more explicitly state safety properties.

• Add documented output schemas for all tools. For one_search, document that it returns {results: [{url: string, title: string, description: string, ...}]}. For one_scrape, document the structure of screenshot data, markdown content, etc. For one_map and one_extract, document the returned URLs and content blocks.
• Enhance error messages with recovery hints. Instead of returning raw error text, parse error types and return guidance: 'Search failed: too many requests. Try reducing limit or waiting 60 seconds.' or 'URL unreachable. Check that the domain is public and not behind authentication.'
• Expand tool descriptions to include when NOT to use a tool. E.g., 'Use one_scrape for single-page detail extraction; use one_map for bulk URL discovery from a starting page.' This helps LLMs avoid redundant calls.
• Add explicit return-value documentation to the description of each tool. E.g., 'Returns an array of search results with url, title, and description fields.' This gives LLMs an idea of what to expect before calling.
• Consider adding idempotency hints to tool descriptions. The search and scrape tools appear read-only, but this is not explicitly stated in descriptions. Adding 'This tool is read-only and safe to retry' clarifies idempotency expectations.
• Document parameter relationships and constraints more explicitly. E.g., for one_scrape, clarify that 'executeJavascript' action requires 'allowExecuteJavascript: true' and that standard actions are bounded by a timeout.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry