Single tool 'search_web' has a clear action verb and basic schema with Zod validation. However, the description (127 chars) is generic and lacks context on when to use it vs alternatives. Parameters lack detailed constraints (e.g., num range, query format). Output is raw JSON dump with no field documentation, making it hard for LLMs to extract relevant data. Error handling is minimal, only catches exceptions without recovery guidance. No pagination support despite potentially large result sets.

• Document output schema with field descriptions. E.g., 'Returns {organic_results: [{title, link, snippet, position}], answer_box?: {answer, source}, related_searches: [string]}'. This lets LLMs extract relevant fields without parsing raw JSON.
• Add constraints to 'num' parameter: 'Number of results to return (1-100, default: 10)'. Validate input and reject out-of-range values with actionable error.
• Implement pagination: add 'start' (offset) parameter and return 'total_results' and 'has_more' fields. Cap default results at 10-20 to prevent context bloat.
• Enhance error handling with recovery guidance. E.g., 'API rate limit exceeded. Wait 60 seconds before retrying.' or 'Invalid query format. Ensure query is a non-empty string.'
• Add tool description context: 'Use this to find recent web information, news, or external references. Not suitable for internal company data, use search_internal_docs instead.' This helps LLMs choose the right tool.