Four tools with complete input schemas and descriptions. Naming follows verb_noun pattern (search_items, get_item, get_seller, get_seller_items). Descriptions are adequate (100-200 chars) and explain WHAT and WHEN. Parameters are typed with enums for constrained inputs (sort, status, condition, format). However, output schemas are not documented, LLMs cannot predict response structure. Error handling is absent; tools return raw API responses without recovery guidance. No tool annotations (readOnlyHint, idempotentHint). Parameter descriptions lack format/range details (e.g., 'Minimum price in JPY' doesn't specify if negative values are rejected). Missing pagination documentation for search_items despite accepting pageToken.

• Document output schemas for all tools. Example: search_items returns {items: [{id, title, price, seller_id, ...}], nextPageToken: string}. This enables LLMs to chain tools and extract required fields.
• Add error handling with recovery guidance. Example: 'Item not found. Try search_items() to find the correct item ID.' or 'Invalid seller ID format. Seller IDs are 10-digit numbers.'
• Add tool annotations via inputSchema metadata. Mark all tools with readOnlyHint: true (they only read Mercari data, never modify). This signals to LLMs that retries are safe.
• Expand parameter descriptions with constraints. Change 'Minimum price in JPY' to 'Minimum price in JPY (0-9999999; omit for no minimum)'. Change 'How many results to return' to 'Number of results per page (1-120, default 20).'
• Document pagination explicitly. Add to search_items description: 'Returns up to `limit` results. Use nextPageToken from response to fetch the next page. Pagination is required for queries with >20 results.'