describe_tariff_system
ChatGPTReturn the native profile and exact guide for a tariff system, optionally selecting a version instead of the active runtime version.
Overview
Tariff Code Compliance gives Claude read-only access to current tariff systems, exact versions, classification nodes, and graph evidence. Claude can ask for missing product facts, compare plausible HS and HTS codes, and explain the evidence, rejected alternatives, assumptions, and uncertainty behind a recommendation. The result is non-binding research support, not legal advice or a binding customs ruling; the user decides whether and how to use it.
Tools
get_customs_bulletin_document
ChatGPTRead saved, prepared CBP Customs Bulletin text independently of search. Pass the search result issue_path, research as_of and version on the first call. For every continuation keep those three fields unchanged and set only page_token to the exact token returned by the preceding page. Reports preparing, failed or explicitly stale evidence; never downloads or parses PDFs during the request.
get_node
ChatGPTFetch one native tariff node by its opaque node_id.
list_node_edges
ChatGPTList filtered native incoming and outgoing edges for an opaque node_id.
list_tariff_systems
ChatGPTList enabled tariff systems and the explicit versions available for native graph traversal.
render_research_panel
ChatGPTUpdate the compact research companion after meaningful research steps and chat answers. Chat owns all clarification: do not interview in the panel or repeat its candidate table in chat. Preserve known product facts, up to four milestones, and up to three researched candidates with reasons and leading/alternative/set-aside states. Include one unresolved next step. Use researched node IDs in one exact tariff version; optionally include up to two researched CBP ruling numbers and comparisons. The server retrieves source text and links for on-demand inspection, distinct from model interpretation. No invented source text or URLs. Empty arguments open the available default tariff system. No classification, duty calculation, or account mutation is performed.
resolve_nodes
ChatGPTResolve native tariff nodes by source key, one or more typed identifiers, and optional version, kind, and effective-time filters.
search_cross_rulings
ChatGPTSearch the selected immutable retained CROSS snapshot lexically, retrieve one exact retained ruling with complete evidence-version-bound text continuations, or find rulings containing normalized legal and tariff references. Lexical search accepts quoted phrases and uppercase AND, OR, NOT, and parentheses, plus optional collection, ruling-date, category, and normalized-reference filters with relevance or explicit date ordering. Results include ruling date, categories, tariff numbers, exact source offsets, prepared-evidence version, source capture availability, direct lifecycle relationships, and snapshot coverage. Empty or stale retained evidence is an explicit research limit; no request-time CBP fallback is attempted.
search_customs_bulletins
ChatGPTSearch prepared CBP Customs Bulletin evidence with required publication dates, normalized phrase or all-term matching, and exact ruling identifiers. The response echoes the supplied tool_input, reports the selected immutable index_scope and concrete unavailable_sources once, and may return useful results despite source gaps. Follow every page while has_more is true, including empty intermediate pages, by repeating the original input with next_page_token as page_token. A mention or empty results array is not a legal-status determination.
search_hts
ChatGPTSearch the exact selected US HTS build with SQLite full-text search across provisions, headings, ancestor context, and indexed notes. Optionally restrict before ranking with scope={type:general_note|gri|chapter,number:string}; use number:"" with general_note or chapter to search that complete note family, and combine chapter: with kinds=["note"] for all chapter and section notes. Scope intersects with kinds. Results retain stable graph node IDs and official evidence locators for graph navigation. Use semantic_search_hts separately for conceptual discovery.
search_nodes
ChatGPTSearch the exact selected US HTS build's lexical index for classification nodes by HTS code prefix or related tariff term with kinds=["classification"], locales=["en"], and no effective_at. Results retain stable graph node IDs for get_node and graph navigation; optionally select a version instead of the active runtime version.
semantic_search_cross_rulings
ChatGPTFind conceptually related ruling-body passages in the selected immutable CROSS snapshot with optional collection, ruling-date, category, and normalized-reference filters applied before ranking. Results include ruling date, categories, tariff numbers, exact source offsets, and similarity. The server supplies the embedding; semantic failure never falls back to lexical search. Similarity and empty results are evidence-discovery outcomes, not legal determinations.
semantic_search_customs_bulletins
ChatGPTFind conceptually related passages in a prepared CBP Customs Bulletin snapshot. The response echoes the supplied tool_input and shares the lexical result and coverage contract, including selected immutable index_scope and concrete unavailable_sources. Global searches return at most one passage per issue for diversity; an exact issue path may return several. Supports optional inclusive publication dates, returns 1-20 top passages with raw cosine similarity and exact locators, has no pagination, and never falls back to lexical search. Similarity or empty results is not a legal-status determination.
semantic_search_hts
ChatGPTFind conceptually related provisions and notes in the exact selected US HTS build using text-embedding-3-large and S3 Vectors. Optionally restrict before ranking with scope={type:general_note|gri|chapter,number:string}; use number:"" with general_note or chapter to search that complete note family, and combine chapter: with kinds=["note"] for all chapter and section notes. Scope intersects with kinds. Results are hydrated from the matching SQLite artifact and retain stable graph node IDs and evidence locators; semantic similarity does not replace legal analysis or graph navigation.
traverse_nodes
ChatGPTTraverse the bounded native graph from one or more opaque node_ids.
