How AI agents can classify products and compute landed cost

Classification and duty calculation are exactly the kind of task AI agents should own: structured inputs, a fixed schema of rules, and an answer that has to be sourced rather than guessed. The catch is that a language model on its own will happily invent an HTS code and a plausible-looking duty rate — which is worse than useless when a wrong number means a CBP penalty. PortRobin is built so agents get the reasoning from AI and the rate from a schedule of record.

The two-part problem

Getting a landed cost has two distinct steps, and they have different trust models. Classification — turning "men's cotton hoodie" into 6110.20.2079 — is a judgment task where AI genuinely helps, provided it's grounded in the real heading texts and the General Rules of Interpretation. Rating — turning that code plus an origin into a duty figure — is a lookup task where the only acceptable source is the official schedule. PortRobin splits them deliberately.

Classifying a product via the API

Send a product description (and optionally an origin) to the classification endpoint. You get back candidate HTS lines ranked by confidence, each with the heading text, a plain-English gloss, and the reasoning for the match. Your agent can then present the top candidate, ask a clarifying question, or fall back to a broader heading — the same workflow a human classifier follows, just faster.

Computing landed cost via the API

Pass the resolved HTS code, the customs value, the origin and the freight/insurance figures to the landed-cost endpoint. It resolves the base rate, applies the relevant Section 301 / 232 and other Chapter 99 measures for the origin, adds MPF and HMF, and returns an itemised breakdown — every line tagged with its source and verification date. If a rate can't be sourced, the response says so rather than smoothing over the gap.

Using the MCP server

For agent frameworks that speak the Model Context Protocol, PortRobin exposes the same capabilities as MCP tools — classify a product, look up a code, compute landed cost — so an AI agent can call them natively inside a conversation or workflow. The tool results carry the same sourced-or-null contract as the REST API, which means an agent can reason about whether it has a trustworthy number or needs to tell the user to verify.

• classify — description in, ranked HTS candidates with reasoning out.
• lookup — an HTS code in, its official description, rate and measures out.
• landed_cost — code, value and origin in, an itemised sourced breakdown out.
• Every tool returns a source URL and date, or an explicit "verify" state.

Why the honesty contract matters for agents

An agent is only as good as the worst number it will confidently repeat. By returning null instead of a guess when a rate isn't verified, PortRobin lets your agent do the right thing: flag the uncertainty, suggest a binding ruling, or route to a human. That's the difference between an agent that helps an importer and one that gets them fined.