BuyWhere MCP has solid naming conventions (all tools start with action verbs: search_, get_, compare_) and complete input schemas with type definitions. Descriptions are present and reasonably detailed (avg ~150 chars). However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Parameter descriptions are good but lack explicit constraints (e.g., no enum for country codes, no min/max for limit). Error handling is basic (McpError wrapping) with no recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite all tools being read-only. Composition is clean, each tool has one responsibility and tools chain well (search_products → get_product → get_price → get_affiliate_link).

• Document output schemas for all tools. Add a 'returns' field to each tool definition showing the structure of formatProduct() and formatListing() output (e.g., {product_id: string, title: string, price: {amount: number, currency: string}, ...}).
• Add tool annotations: set readOnlyHint=true and idempotentHint=true on all 6 tools to signal safety to agents.
• Formalize parameter constraints in JSON Schema: add enum for 'country' (e.g., ['sg', 'my', 'id', 'th', 'vn']), add minItems=2 and maxItems=5 to compare_prices.product_ids schema, add minimum=1 and maximum=50 to search_products.limit schema.
• Enhance error messages with recovery guidance. When get_product fails with 404, return: 'Product not found. Try search_products(query="...") to find the correct product_id.' When API rate limit is hit, return: 'Rate limited. Retry in 60 seconds.'
• Add pagination support to search_products: introduce 'offset' (default 0) and 'limit' (default 10, max 50) parameters, and return {results: [...], total_count: number, has_more: boolean} to enable multi-page discovery.