Your Amazon business, run from your own AI.
Connect Seller Central once. Claude or ChatGPT reads the profit Amazon actually posted to your account, drafts the bid, budget and listing changes, and waits for you to approve them.
- Numbers from your settlement, not an estimate
- Nothing reaches Amazon without your approval
- Free plan, no card
Running on reports: Five tabs. One question. Last week's answer.
Seller Central shows sales. The ads console shows ACoS. The fee is in a settlement report you download on Friday. Whether a campaign actually made money is a spreadsheet nobody has time to build.
"Which campaign lost money last month?" → an evening with CSV exports.
Running with your own AI: One question. The measured answer. Then a draft you approve.
AgentXray reads the settlement line items Amazon posted — referral, FBA, storage, promotions — per unit, per campaign. Your assistant asks, gets the number with its basis and its gaps, and files a change into your approval queue.
"Pause the arm that's losing money." → a draft, validated against Amazon, waiting for your yes.
What you can ask on day one
Each line is a real tool on the connector. Answers carry where the number came from and what it does not include.
- How did today go?
- What did I actually make in the last 30 days — and what is not included in that number?
- Which of my SKUs is losing money right now?
- What is my break-even ACoS on this product, from my own fees?
- Which search terms spent money and never converted?
- Where am I losing the Buy Box, and by how much?
- Is any FBA stock heading toward long-term storage fees?
- Did the bid change from two weeks ago actually help?
Measure → draft → ask Amazon → you approve → verify
Dashboards stop at the first step. The connector runs all five: it measures from your settlement, writes the change for you, validates it against Amazon before it goes live, waits for your approval in the browser, and judges the result fourteen days later on net contribution — or says the data is too noisy to tell.
Who this is for
One-person brand
You are the founder, the PPC manager and the accountant. You want one number you can trust and a change you can approve from your phone.
Growing private label
Ten to a few hundred SKUs across US, CA, UK, DE or AU. Per-SKU contribution and per-campaign net matter more than a blended ACoS.
Sellers who already use Claude or ChatGPT
You have the assistant. It just cannot see your account. The connector gives it the numbers and a way to draft, never a way to act alone.
What we do not do — said up front
- We do not estimate fees. If a settlement has not posted yet, the figure is shown as a gap, not a guess.
- We do not apply changes automatically. Every draft is approved by a person in the browser; the connector token is refused at the approval endpoint.
- We do not run product research. The market-data subscription behind the research tools is paused; those tools are labelled as such.
- We have not commissioned a third-party penetration test, and we say so on the security page.
Questions
Which marketplaces?
US, CA, UK, DE and AU — the ones the measurement layer runs on today. The fees come from your own settlements, so nothing is region-locked; coverage grows as marketplaces are walked end to end.
Do I need Claude or ChatGPT?
No. The panel shows the same measured numbers and the same approval queue. The connector is the part that lets an assistant you already pay for ask the questions.
Can the AI change my account?
It can draft. Bids, budgets, negatives, campaign state, listing edits and shipments are filed as proposals. Only a person approving in the browser makes them reach Amazon.
What does it cost?
A free plan that shows what a scan returns before you connect anything, then paid tiers that differ only by monthly order volume. Every feature is in every paid tier; AI questions are never metered.