Three tools with complete input schemas and descriptions, but descriptions are generic and lack LLM-optimized guidance. All tools are read-only search operations with similar structure. Parameters are typed and constrained (count 1-50, market codes, freshness enum), but descriptions lack actionable context about when to use each tool vs. alternatives. No output schema documentation. Error handling returns plain strings without recovery guidance. Rate limiting is implemented server-side but not exposed to the agent.

• Convert plain-text responses to structured JSON: {results: [{title, url, description, date?, provider?}], total_count, next_offset}. This enables LLMs to extract and chain data reliably.
• Enhance tool descriptions with decision trees: 'Use bing_web_search for general information (recipes, definitions, websites). Use bing_news_search for current events and recent developments. Use bing_image_search for visual content only.' Add 50-100 chars of context per tool.
• Implement error categorization: return {error: string, category: 'retryable'|'user_fixable'|'fatal', suggestion: string}. E.g., 'API key missing' → category: user_fixable, suggestion: 'Set BING_API_KEY environment variable.'
• Add enum constraint to freshness parameter: {type: 'string', enum: ['Day', 'Week', 'Month']} to prevent hallucinated values.
• Document result limits in descriptions: 'Returns up to 50 results per request. Use offset parameter for pagination.' Clarify that count is capped at 50 even if requested higher.