browse-mcp provides three well-named tools with clear, action-verb prefixes (web_search, web_fetch, web_download). Tool descriptions are substantive (160-220 chars) and explain use cases and constraints. Input schemas are present with typed parameters and descriptions. However, output schemas are not documented, the source shows tool registration but no return type specifications. Error handling guidance is minimal. The server demonstrates good foundational quality with room for improvement in output documentation and error recovery patterns.

• Add formal output schemas for each tool. Document what fields web_search returns (e.g. results array with {title, url, snippet, rank}), what web_fetch returns (e.g. text content, metadata, file path for large responses), and what web_download returns (bytes_written, final_path, file_size). This enables LLMs to chain tools and extract correct fields.
• Implement error recovery patterns. For invalid URLs in web_fetch, return 'Invalid URL format. Must start with http:// or https://. Example: https://example.com/page'. For timeout errors, return 'Request timeout after {timeout}s. Try reducing timeout parameter or retrying after a delay.' For network errors, indicate 'Retryable: Failed to reach {url}. Check connectivity and retry.' This enables the LLM to self-correct.
• Add enum constraint to web_search provider parameter. Instead of a free-form string, declare: provider (enum: 'duckduckgo' | 'tavily' | 'serper', default: 'duckduckgo'). This prevents the LLM from inventing provider names.
• Document constraints for numeric parameters. Add to web_search max_results: 'Between 1 and 20; default 10. Larger values increase latency and response size.' Add to web_fetch and web_download timeout: 'Between 1 and 120 seconds; default 30 for web_fetch, 120 for web_download. Shorter timeouts reduce wait time but risk incomplete transfers.'
• Add pagination hints to web_search. Clarify: 'Returns up to max_results items per call. No offset/cursor is currently supported, if you need different results, refine the query. Consider using multiple specific queries to explore different aspects of the topic.'