Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Tool names follow verb_noun convention (googleSearch*), which is good. However, descriptions are minimal (under 100 chars), parameter descriptions lack detail about format/constraints, and output schemas are not formally documented. Error handling returns JSON strings rather than structured MCP responses. No tool annotations (readOnlyHint, etc.) despite READ_ONLY risk classification.

• Add formal output schema documentation. Define the structure returned by each tool (e.g., {engine: string, query: string, total_results: string, items: [{title, link, snippet}]}). Return structured MCP content objects instead of JSON strings.
• Expand tool descriptions to 150-250 chars. Include WHAT (searches Google), WHEN (when you need current web results), and WHAT IT RETURNS (top 5 results with title, link, snippet). Add prerequisite note: 'Requires GOOGLE_API_KEY and GOOGLE_CX_ID environment variables.'
• Add readOnlyHint: true to both tool schemas to signal read-only safety. Per pattern:tool, annotations help agents reason about side effects.
• Document parameter constraints in descriptions. For query: 'Non-empty string, max 2048 chars, supports Google search operators (site:, filetype:, etc.)'. For siteFilter: 'Domain name (e.g., example.com) or URL prefix (e.g., example.com/blog). Restricts results to specified site.'
• Implement per-item error reporting. If API returns 403 (quota exceeded), return structured error with retry guidance: {error: 'API quota exceeded', retryable: true, nextRetryAfter: 3600}. Per pattern:recovery-guide, errors should guide agent next steps.