Three well-defined tools with explicit schema registration, clear descriptions, and good parameter documentation. The server follows verb_noun naming patterns and includes helpful context for LLM selection. However, there are gaps in output schema documentation, error recovery guidance, and some parameter descriptions could be more specific. The schemas use Zod validation and are properly registered via server.registerTool(). Descriptions are detailed enough to guide LLM tool selection but lack explicit output field documentation. All three tools are READ_ONLY with no destructive operations, reducing error handling complexity.

• Define and document complete output schemas for all three tools. Include return type definitions with all fields that will appear in the response (e.g., for get_product: { asin: string, title: string, price: number, mrp: number, discount_percent: number, rating: number, review_count: number, availability: string, brand: string, seller: string, delivery_info: string, keepa_url: string }). This enables LLM planning and composition.
• Enhance error messages to indicate retry strategy. Examples: 'BotBlockedError: Amazon served a CAPTCHA. [RETRYABLE] Wait 60 seconds and try again with a narrower query.' 'FetchFailedError: Network timeout [RETRYABLE] Retry in 10 seconds.' 'Invalid ASIN format [USER_FIXABLE] Use a 10-character ASIN like B0123456789 or a full amazon.in URL.'
• Explicitly state parameter constraints in descriptions using formatted text. For asin_or_url, change description to: 'Amazon.in ASIN (exactly 10 alphanumeric characters, e.g. B0A1B2C3D4) OR a full product URL containing /dp/<ASIN>. URL example: https://www.amazon.in/dp/B0A1B2C3D4'
• Document the response structure for search_amazon_in. Include field names: listings (array of {asin, title, price, rating, review_count, availability, ...}), cheapest_in_stock (single item or null), best_value (single item or null), query_matched (count or total_found), page_info (current_page, has_next_page).
• Add total_results or has_more field to search_amazon_in output so the LLM knows whether additional pages exist without trial-and-error pagination.