Server has 5 tools with complete input schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (find_goods, get_item_info, create_tpwd). Descriptions are present but vary in clarity, some include scope/API IDs (27939, 16189) that are implementation details, not user-facing context. Parameters have types and defaults, but lack detailed constraints (e.g., 'count' has no min/max bounds despite API limits). Output is unstructured text (formatted strings) rather than structured JSON, forcing LLMs to parse prose. Error handling returns plain strings without recovery guidance. No tool annotations (readOnlyHint, idempotentHint). Missing output schema documentation.

• Refactor all tool responses to return structured JSON: {success: bool, data: {...}, error?: string, total_count?: int, next_cursor?: string}. Parse internal API responses and reshape to match agent data model.
• Add min/max constraints to numeric parameters in schema and description. E.g., count: {type: integer, minimum: 1, maximum: 100, default: 20} + description: 'Results per page, 1 - 100, default 20'.
• Remove Scope IDs and API method names from descriptions. Rewrite to user intent: 'Search Taobao products by keyword and get affiliate links' instead of 'Search Taobao products with affiliate links. Scope: 27939 (upgrade search API)'.
• Document output schema for each tool. Include field names, types, and whether fields are always present. E.g., find_goods returns {items: [{id: string, title: string, price: string, shop: string, url: string}], total_count: int}.
• Enhance error messages with recovery steps. E.g., 'Missing TAOBAO_APP_KEY environment variable. Set it before retrying.' or 'Invalid material_id. Call recommend_products with a valid ID from featured_products first.'