Analysis Prompts

Fifteen standardized tariff analysis workflows. The tools give you the schedule; the analyses give you the method for using it.

Each one orchestrates the HTS MCP tools into a reproducible analysis with a consistent output shape: a status, a confidence score, the evidence anchors behind every conclusion, and an explicit list of what it could not determine.

How they work

These are exposed as MCP prompts, not tools. That difference governs how you use them:

• You invoke them, the model does not. MCP prompts are user-controlled. A connected model can see the full menu by calling the list_analyses tool, and can tell you what is available, but it cannot start an analysis on its own. You select one.
• They appear as a slash command namespaced to the server. The exact rendering differs between MCP clients, so the prompt name in the table below is the reliable identifier.
• The server returns the method, not the answer. Selecting an analysis returns roughly 10,000 to 15,000 characters of workflow instructions: which tools to call, in what order, how to weigh the evidence, and the required output schema. Your model then executes it against the live HTS tools. No language model runs on our side.

That last point is the design: conclusions are produced by a model you control, from data we serve and can cite, with the reasoning steps written down in advance rather than improvised per request.

Data tiers

Every analysis declares how much of its workflow current data can support.

Tier 3 analyses return the full rigorous workflow plus a precise list of what they cannot source. Use them as structured method and a data-request checklist, not as a finished determination.

Available analyses

Required arguments are enforced. Omitting one returns an error rather than a partial result.

Tier 1

Evidence-backed classification with GRI application, legal notes, CROSS rulings and rejected alternatives. Duty calculation with rate parsing, MPF and HMF, effective rates and sensitivity. Exposure quantification with ranked mitigation options carrying dollar figures.

Tier 2

Transaction value determination with statutory additions and related-party assessment. Section 301, AD/CVD and Section 232 screening with rate stacking. Batch monitoring for rate changes, remedy actions and HTS restructuring.

Tier 3

Discovering them at runtime

list_analyses is a regular tool, so a connected model can call it to see everything available:

{
  "analyses": [
    {
      "prompt": "classify",
      "name": "Defensible HTS Classification & Precedent Report",
      "tier": 1,
      "description": "Evidence-backed classification with GRI application, notes, CROSS rulings, and alternatives"
    }
  ],
  "total": 15
}

At the protocol level, prompts/list returns the same fifteen with their argument schemas.

Output shape

Every analysis conforms to a shared structure, so results are comparable across types and can feed each other:

Missing inputs downgrade confidence and generate a data-request checklist. They do not produce a guess.

Chaining

Classification is the prerequisite for most other analyses: a wrong HTS code invalidates the origin rule, the remedy screening and the duty calculation downstream of it. A typical new-supplier sequence runs classification, then valuation, then origin, then landed cost, then remedy screening, then sourcing strategy.