AI agent workflows

Ask Claude, ChatGPT, Gemini how to get international trade work done with Tandom.

A broker, importer, or forwarder should not need to know every endpoint name before seeing value. They can describe the task in plain English; an AI agent can read the Tandom docs, pick the API functions, and produce the calls.

The agent pattern

1. User describes a trade-ops task.
2. Agent reads Tandom's LLM docs or MCP tool list.
3. Agent maps the task to API functions.
4. Operator gets a cited, auditable result.

Give the agent Tandom's docs

Point it at the LLM-readable reference or paste the full Markdown dump into the chat.

https://docs.tandom.ai/llms-full.txt

Ask from the operator's point of view

Describe the actual customs task, the systems you use, and what result would be useful back in your workflow.

I am a customs broker preparing entry lines from invoices. Which Tandom API calls should I use?

Let it choose REST or MCP

The same API can be called directly over HTTP, through generated code, or through the MCP endpoint for agent tools.

POST https://mcp.tandom.ai/mcp

Copy this into any agent

A generic prompt for operator-led discovery

This works in Claude, Codex, ChatGPT, Cursor, or any tool that can fetch public URLs. For execution, give it a test key or connect it through Tandom's canonical MCP endpoint.

You are helping a customs, importing, or freight-forwarding operator decide whether Tandom can automate a task.

Use the Tandom API docs:
https://docs.tandom.ai/llms-full.txt

Task:
<paste the user's real workflow here>

Return:
1. The Tandom API functions to call, in order.
2. Which inputs I already have and which inputs are missing.
3. A sample REST or MCP call with realistic parameters.
4. What the response should be used for in my workflow.
5. Any available-now vs closed-beta boundary.

Scenario library

What the user asks, and what the agent should call

Validated

5 of 5 scenarios run end-to-end on 2026-05-04 against Claude (Opus 4.7) with the Tandom MCP connector. Scenario #5 also exercised the Composio (Google Sheets) custom MCP connector to prove multi-MCP composition. Each card below carries its own Validated date and a one-paragraph summary of what Tandom surfaced in that run.

See /api/connectors for the install steps used in these tests.

Customs broker

Turn one invoice line into an auditable entry-line decision.

Validated 2026-05-04

User asks

I have an invoice line for stainless steel forged flanges from China, HTS hint 7307.91.50.10, value $10,000, entry date April 30 2026. Show me which Tandom API functions would calculate duties, flag AD/CVD, identify PGA requirements, and produce evidence I can keep with the file.

Agent plan

1Normalize the HTS code and pull hierarchy / notes if the code needs review.
2Calculate the full duty stack as of the entry date.
3Run AD/CVD scope analysis with product description, country, and manufacturer if available.
4Check PGA flags and return the required agency documents.
5Package citations and warnings for the broker's entry packet.

API functions

GET/v1/hts/hierarchyAvailable now
GET/v1/duty/calculateAvailable now
GET/v1/adcvd/checkAvailable now
GET/v1/pga/checkAvailable now
POST/v1/forms/cbp-7501Closed beta

Output

A line-level duty stack with MFN, Chapter 99, AD/CVD advisory, MPF/HMF, source citations, PGA flags, and a path to 7501 generation when the forms API is enabled.

Status boundary

The intelligence calls are available now. CBP form generation is closed beta.

Validation run · 2026-05-04

Client: Claude (Opus 4.7) via the Tandom MCP connector

Tandom returned the duty stack matching expected $8,097.14 (5.6% MFN + 50% Section 232 derivative + 25% Section 301 List 3 + MPF + HMF). Claude flagged the iron-vs-stainless HTS misclassification on first pass, ran the AD/CVD check against the corrected 7307.21 branch, and surfaced A-570-064 (257.11%) + C-570-065 (174.73%) — i.e., ~$51K properly-filed exposure vs ~$8K wrong-HTS. Closed with a defensible refuse-to-file recommendation absent mill cert and manufacturer ID.

Freight forwarder operator

Quote a shipment without missing a hidden AD/CVD exposure.

Validated 2026-05-04

User asks

A customer asks us to quote imported aluminum extrusions from Vietnam. We have a rough product description and a possible HTS heading, but the shipper is not sure about manufacturer details. Can Tandom help an AI agent decide what to ask for and what landed-duty risk to surface before we quote?

Agent plan

1Search HTS candidates from the product phrase and collect the notes the operator should verify.
2Ask for missing manufacturer / exporter details only if they materially affect AD/CVD rates.
3Run AD/CVD check with the best HTS candidate and country.
4Calculate duties for quote scenarios once value and quantity are known.
5Return quote warnings instead of a false all-clear when scope or manufacturer is ambiguous.

API functions

GET/v1/hts/searchAvailable now
GET/v1/hts/notesAvailable now
GET/v1/adcvd/checkAvailable now
GET/v1/duty/calculateAvailable now
POST/v1/quote/landed-costClosed beta

Output

A quote-ready risk memo: likely HTS candidates, AD/CVD cases to investigate, rate range, missing data questions, and landed-duty assumptions.

Status boundary

Duty, HTS, and AD/CVD are available now. Full landed-cost quote orchestration is closed beta.

Validation run · 2026-05-04

Client: Claude (Opus 4.7) via the Tandom MCP connector

Decision-support flow without a concrete HTS — Claude correctly framed the workflow rather than running tools blindly. Named every tool by its current wire name (tandom_hts_search, tandom_adcvd_check, tandom_chapter99_applicable, tandom_pga_check, tandom_duty_calculate, tandom_regulatory_status), surfaced the Vietnam-aluminum circumvention angle, demanded producer/exporter identity and country-of-melt-and-pour as gating facts before a firm quote, and offered to run the full sequence once the user hands over a description and HTS heading.

Importer compliance team

Monitor a product catalog for tariff and AD/CVD changes.

Validated 2026-05-04

User asks

We import 180 SKUs from China, India, and Mexico. Some already have HTS codes. I want ChatGPT or Codex to tell me how to use Tandom to find current exposure and set up alerts when duties, AD/CVD cases, or Chapter 99 tariffs change.

Agent plan

1Split the catalog into known-code and unknown-code rows.
2Use HTS search only where codes are missing or suspicious.
3Resolve current duty and AD/CVD exposure for each code / country pair.
4Pull AD/CVD order metadata for the watch list.
5Register the catalog for alerts once monitoring access is enabled.

API functions

GET/v1/hts/searchAvailable now
GET/v1/chapter99/applicableAvailable now
GET/v1/duty/calculateAvailable now
GET/v1/adcvd/ordersAvailable now
POST/v1/monitor/catalogClosed beta

Output

A SKU exposure table with duty rates, active AD/CVD orders, citations, uncertainty flags, and a monitoring plan for future alerts.

Status boundary

Current exposure lookup is available now. Catalog monitoring and webhooks are closed beta.

Validation run · 2026-05-04

Client: Claude (Opus 4.7) via the Tandom MCP connector

Claude ran the full tool sequence on a worked SKU (7307.19.30.85, CN, $5,000) and produced a downloadable markdown deliverable (tandom_catalog_exposure_playbook.md) — splitting 180 SKUs into six origin × sector buckets, defining a six-tool deterministic sequence with top-30 / next-90 / remaining-60 prioritization, and a 21-column exposure-table format. Strongest finding: Tandom's scope analysis flagged two heading-level AD/CVD matches (A-570-875 75.5%, A-570-881 111.36%) but correctly determined them likely out of scope at 85–92% confidence (cast/malleable iron orders vs steel product) — a 186% phantom exposure that a naïve heading match would have produced. Also surfaced the cbp_ace_messages dataset error state for transparent data-health flagging.

Importer or broker

Prepare a reasonable-care classification report.

Validated 2026-05-04

User asks

I need a defensible classification report for a new consumer electronics product. Ask me for the minimum missing facts, then use Tandom API functions to classify it, cite CROSS rulings and HTS notes, calculate duty, and create a report my team can review.

Agent plan

1Ask for product composition, principal use, value, origin, and technical specs.
2Search HTS terms and retrieve notes before proposing a 10-digit code.
3Run classification and report generation when the beta endpoint is enabled.
4Calculate duty and Chapter 99 exposure for the selected code.
5Return a review packet with confidence, alternatives, citations, and unresolved assumptions.

API functions

GET/v1/hts/searchAvailable now
GET/v1/hts/notesAvailable now
POST/v1/classifyClosed beta
POST/v1/classify/reportClosed beta
GET/v1/duty/calculateAvailable now

Output

A classification workup with the proposed code, GRI reasoning, citations, duty result, alternative codes, confidence, and review questions.

Status boundary

Reference and duty calls are available now. AI classification and report generation are closed beta.

Validation run · 2026-05-04

Client: Claude (Opus 4.7) via the Tandom MCP connector

Claude refused to classify until it had the GRI-relevant facts: principal function decomposition, battery chemistry/Wh, connectivity, country of origin and export, manufacturer/exporter identity, customs value per unit and annual volume, steel/aluminum derivative content, and intended end use. After receiving the brief (battery-powered handheld Bluetooth speaker, plastic + metal grille, USB-C, made in Vietnam), Claude ran the Tandom tool chain, identified IEEPA reciprocal exclusion 9903.01.32 as suppressing Vietnam's 20% reciprocal tariff (only Section 122 at 10% applies), flagged FCC certification, and produced a downloadable Word deliverable (Classification_report_bluetooth_speaker_vn.docx).

Importer compliance lead

Use a Google Sheet as the SKU case tracker and let Claude refresh exposure data without leaving the spreadsheet.

Validated 2026-05-04

User asks

I have a Google Sheet of 40 active SKUs with HTS, country, value, and a last-checked date. Refresh anything older than 7 days: re-run Tandom duty, AD/CVD, and Chapter 99 checks, write the updated rate, scope determination, and effective rate back to the row, and add a flag if any rate moved by more than 1 percentage point since the last check.

Agent plan

1Read the SKU sheet via the Google Sheets connector and filter rows where last_checked is older than 7 days.
2For each stale row, run Tandom checks in parallel (chapter99 applicability, AD/CVD scope, duty calculation).
3Diff the new rates against the stored values and flag any movement greater than 1 percentage point.
4Write effective rate, scope determination, AD/CVD rate, alert flag, and a verification timestamp back into the row.
5Return a delta summary listing which SKUs changed and why so the compliance lead can review before alerting downstream teams.

API functions

GET/v1/chapter99/applicableAvailable now
GET/v1/adcvd/checkAvailable now
GET/v1/duty/calculateAvailable now
GET/v1/regulatory-data/statusAvailable now

Output

An updated case tracker sheet plus a delta memo summarizing which SKUs moved, the new vs. previous rates, and the suppression / scope reasoning behind any change.

Status boundary

Tandom intelligence is available now. Anthropic's curated directory does not yet include a standalone Google Sheets connector — for this workflow we use the Composio hosted MCP (mcp.composio.dev) which brokers Google Sheets read/write under your Google account. Install it as a custom connector alongside Tandom.

Validation run · 2026-05-04

Client: Claude (Opus 4.7) with both the Tandom MCP connector and Composio (Google Sheets) custom connector enabled

Full multi-MCP composition end-to-end. Claude seeded the Sheet (header + 5 rows), fired 15 Tandom calls in parallel (5 SKUs × chapter99/adcvd/duty), wrote columns H–M back into each row (new_effective_rate, delta_pp, adcvd_status, chapter99_active, alert_flag, last_refreshed), and returned a delta memo. Strongest finding: SKU-5 (MX wire harnesses) prev 0% → new 30% because Section 232 Auto Parts (9903.94.05 +25%) lands on top of MFN even with USMCA-origin claim — Claude correctly noted that USMCA does not zero out S232 auto-parts. SKU-1 (CN stainless flanges) flat in published rate but flagged as ~$43K AD/CVD cash-deposit exposure on a $10K customs value. SKU-3 and SKU-4 returned HTS not in DB — flagged as 10-digit rebase candidates rather than papered over.

Agent execution

Expose the same functions as MCP tools

REST is best when you are building an integration. MCP is best when an agent should discover and call tools during a conversation. Tandom's connector path starts with read-only customs intelligence tools, then expands to classification, reporting, forms, and monitoring as those product surfaces leave closed beta.

Open llms.txt →Browse API reference →Install connector →

// JSON-RPC 2.0 — discover Tandom tools
POST https://mcp.tandom.ai/mcp
Authorization: Bearer tk_test_...
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list",
  "params": {}
}

// Then call the selected tool; execution inherits the REST endpoint's limits.

Start with a sandbox key

Test keys return fixture data and avoid real database writes or Claude calls. They are the right first step for letting an agent explore Tandom safely.