Server has 4 tools with basic descriptions and partial parameter schemas. The tools themselves are well-named with action verbs (web_search, fetch_webpage, wikipedia_search, instant_answer) and non-trivial descriptions (60-110 chars), but lack the depth needed for reliable LLM selection and execution. Baseline comparison: tool descriptions average 194 chars (these average ~85 chars); all params need descriptions (these lack depth); 100% of A+ tools have documented return types (this server has zero).

• Document return schemas for all tools. For web_search, explicitly state: 'Returns array of {title (string), url (string), snippet (string), source (string)} objects, plus errors array if any search methods failed.' Include field descriptions and examples.
• Add min/max constraints to numeric parameters. web_search numResults should be: 'Number of results to return (default: 10, minimum: 1, maximum: 50).' Apply min/max to any numeric input.
• Enhance parameter descriptions to 70+ characters with format and constraint details. Example: 'The target webpage URL (http:// or https://, max 2048 chars). Only text/html content is extracted; binary files are rejected.' State failure modes.
• Implement structured error responses. Instead of console.error(), return {error: {code, message, recovery_hint}}. Example: 'All search backends failed. Try again in 30 seconds, or use instant_answer as a fallback.' Categorize errors as retryable vs. fatal.
• Add tool-level composition hints in descriptions. For fetch_webpage: 'Complements web_search(), pass URLs from search results directly into this tool to extract full article text.' For web_search: 'Returns URLs suitable for fetch_webpage().'
• Clarify result caps and filtering. State: 'Results capped at 50 per source. Duplicates are removed. Order reflects source ranking.' If filtering occurs (e.g., adult content), document it.
• Add prerequisites and dependencies to descriptions. Example for wikipedia_search: 'Fastest for encyclopedic queries; slower for current events. Use web_search for real-time results, wikipedia_search for stable reference material.'