カーリル for AI

The three tools target clearly distinct resources and actions: classification metadata (get_ndc_info), library/system-ID lookup (search_libraries), and holdings search (search_books). The descriptions explicitly clarify the dependency between search_libraries and search_books, leaving no realistic misselection risk.

All names use snake_case with a predictable verb_noun structure (get_ndc_info, search_books, search_libraries). The verb choice varies only slightly (get vs search), which is semantically justified and still fully predictable.

Three tools is on the lean side, but each maps to a genuinely separate step of the library-search workflow (identify library, search holdings, interpret classification), so none feel redundant. A couple more tools (e.g. availability) could be added without bloat, hence not a perfect 5.

The surface covers the core read path — find a library, search its holdings, decode NDC classes — but there is no tool for lending/availability status, individual book detail by ID, or pagination/narrowing helpers, which are common needs for this domain. These are notable gaps rather than fatal ones.