Server provides 7 read-only tools with complete JSON schemas and Japanese descriptions. All tools have proper verb_noun naming (search_*, get_*) and structured input schemas with type definitions. However, descriptions are domain-specific (Japanese, EC-CUBE terminology) and lack LLM-optimization guidance (no 'when to use', no prerequisites, no recovery hints). Parameter descriptions are minimal (1-2 sentences). Output schemas are not documented, LLMs cannot predict response structure. No error handling guidance. All tools are READ_ONLY, reducing risk but limiting utility.

• Document output schemas for all tools. Specify response fields, types, and structure. E.g., search_products returns {products: [{id, name, price, url, ...}], total, has_more}.
• Expand descriptions to 100-200 chars with LLM-optimization: state WHAT (e.g., 'Search EC-CUBE products'), WHEN (e.g., 'Use when user asks for products by keyword'), and WHAT IT RETURNS (e.g., 'Returns product list with prices, URLs, and stock status'). Provide English descriptions alongside Japanese.
• Add min/max bounds to numeric parameters. E.g., limit: 1-100 (default 20), offset: ≥0. Add format hints to string params (e.g., category_id is an integer, not a string).
• Add error handling guidance in tool descriptions. E.g., 'Returns empty list if no products match; returns 400 if category_id is invalid (try get_categories first).'
• Implement tool annotations (readOnlyHint=true for all tools) to signal to agents that these are safe, non-destructive calls.