Build an Amazon FBA product research agent

2026-10-26 · 3 min read

Product research is the most time-consuming part of selling on Amazon: filtering thousands of listings, checking competition, estimating fees, and doing the math before ordering inventory. It is also a near-perfect job for an AI agent, as long as the agent has real marketplace data instead of guesses.

This guide builds a research agent in four steps using ASIN402's pay-per-call endpoints. No subscription is involved: the whole pipeline below costs well under a dollar per run.

The workflow

1. Find candidates that match your criteria with the Product Finder.
2. Score each candidate as a selling opportunity.
3. Calculate FBA profit for the survivors using your real unit cost.
4. Report the best options with the reasoning behind them.

Setup

npm install @x402/fetch @x402/evm viem

import { wrapFetchWithPayment, x402Client } from "@x402/fetch";
import { ExactEvmScheme } from "@x402/evm/exact/client";
import { privateKeyToAccount } from "viem/accounts";

const pay = wrapFetchWithPayment(fetch,
  new x402Client().register("eip155:*", new ExactEvmScheme(privateKeyToAccount(process.env.AGENT_KEY))));
const API = "https://asin402.com";
const get = async (path) => (await pay(API + path)).json();
const post = async (path, body) => (await pay(API + path, {
  method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(body),
})).json();

Fund the agent's wallet with a few dollars of USDC on Base. Use a dedicated wallet.

Step 1: Find candidates

The Product Finder filters the whole Amazon catalog in one call. Prices are in cents. This query looks for mid-priced products that sell steadily, have room for a new seller, and where Amazon itself is not competing:

const finder = await post("/api/data/finder", {
  current_BUY_BOX_SHIPPING_gte: 2000,   // buy box at least $20
  current_BUY_BOX_SHIPPING_lte: 6000,   // ...and at most $60
  monthlySold_gte: 300,                 // ~300+ units sold last month
  totalOfferCount_lte: 8,               // 8 sellers or fewer
  buyBoxIsAmazon: false,                // Amazon doesn't hold the buy box
  current_COUNT_REVIEWS_lte: 400,       // beatable review moat
  perPage: 50,
  sort: [["monthlySold", "desc"]],
});
const candidates = finder.asinList.slice(0, 10);

To focus on one category, add "rootCategory": [<id>]. Get the ID with GET /api/data/category-search?term=kitchen. The full list of filters is at /docs/finder, or see Find profitable products with the Product Finder.

Step 2: Score each candidate

/api/sell-score combines demand, competition, Amazon's presence, price stability, review moat and price fit into a 0-100 score, with plain-language risk flags:

const scored = [];
for (const asin of candidates) {
  const s = await get(`/api/sell-score?asin=${asin}`);
  scored.push(s);
}
const shortlist = scored.filter((s) => s.score >= 60 && !s.signals.amazon_is_seller);

A typical result:

{
  "asin": "B0EXAMPLE1",
  "score": 72,
  "rating": "strong",
  "components": { "demand": 93, "competition": 58, "price_stability": 59, "price_fit": 90, "review_moat": 10, "amazon_penalty": 0 },
  "flags": []
}

Flags are worth reading even on high scores: "Large price swings in the last 180 days (price war risk)" or "Out of stock 40% of the last 90 days" change the decision.

Step 3: Calculate FBA profit

Pass your landed cost per unit (product plus inbound shipping):

for (const s of shortlist) {
  const p = await get(`/api/fba-profit?asin=${s.asin}&cost=6.50&shipping=1.20`);
  s.profit = p;
}

The response includes the referral fee, the FBA fulfillment fee, net profit, margin, ROI, the break-even cost, and the maximum landed cost for a 30% ROI, which is the number to bring to a supplier negotiation.

Step 4: Report

Give the shortlist to your model and ask it to explain the trade-offs, or print a table:

console.table(shortlist.map((s) => ({
  asin: s.asin, score: s.score, price: s.signals.current_price,
  monthly_sales: s.signals.estimated_monthly_sales, net_profit: s.profit.net_profit, roi_pct: s.profit.roi_pct,
})));

What a run costs

Failed calls (an ASIN with no data, for example) are never charged.

Tips from real use

• Score fewer, better candidates. Tighten the finder filters until 10-20 products come back; scoring is the expensive step.
• Check consistency, not just the snapshot. Add salesRankDrops90_gte: 300 to the finder to keep products that sold steadily over the last 90 days (each rank drop roughly equals a sale), not just last week.
• Re-run weekly. Markets move; a product that failed last month because Amazon held the buy box can open up.
• Treat scores as a filter, not a verdict. They rank opportunities from marketplace data; supplier quality, IP risk and your own experience still decide.

Next steps

• Find profitable products with the Product Finder: more filter recipes
• Use the same tools in Claude or Cursor, no code needed