S9·Search For Amazon sellers · October 2026

Amazon's free Claude plugin, and what it won't tell you

In September Amazon put Seller Central inside Claude and charged nothing for it. Switch it on. Then notice the two things it will never say to you, and ask yourself why.

On 23 September at Accelerate, Amazon released a Selling Partner plugin that connects Seller Central to Claude. It is free, it is in beta, and if you are a US seller on a Professional plan you can turn it on today. It reads your listings, your stock and your live sales numbers, and it will draft a price change or a restock plan and wait for you to approve it. It is a good piece of work and I would use it.

It is also Amazon's. That is not a slur, it is just the organisation chart. Amazon earns from you in three ways: a cut of everything you sell, the advertising you buy, and the fulfilment you use. A tool built inside that building will happily help you sell more, spend more and ship more. It is never going to open a conversation by suggesting you advertise less or charge more.

So here are four things worth noticing before you lean on it.

1

It will not tell you to spend less on advertising. Not because anyone has been dishonest about it, but because advertising is one of Amazon's largest businesses and nothing inside Amazon is built to shrink it. The ad numbers are not even in the same place: the Selling Partner plugin carries none, and the separate Ads MCP that has been in open beta since February answers in isolation. One knows what you spent, the other knows what sold, and nothing joins the two to the costs that say whether it was worth it.

2

It will carry out a price cut without ever asking whether it was a good idea. Amazon has deliberately kept the plugin out of pricing advice and margin calls. Read that charitably, as them not wanting to be blamed for your P&L, and it is still the case that the one judgement you actually need is the one it is built not to give. A lower price suits Amazon in every direction. It does not always suit you.

3

It only knows today. It reads current state. It holds no record of what your price was in July, when a competitor took the Buy Box off you, or what your listing said before somebody changed it. You can ask it what is happening. You cannot ask it what changed, which is nearly always the question that leads somewhere.

4

Your costs are a single number in a box. You can type a unit cost into Seller Central and it will do arithmetic with it. What it does not carry is freight, duties, prep, VAT treatment, or the fact that your cost per unit was different on the last three shipments. Profit built on one static figure is a figure, not an answer.

What I built instead, and who it answers to

S9 is the system I built for my own business and now set up for other sellers. We create a data warehouse inside your own Google Cloud account, in your name and on your billing. Every night it pulls your Amazon data into it: orders, traffic, search terms, advertising, listings, returns, fees and stock. If you also sell on Shopify or TikTok Shop, those go in alongside. Setup backfills years of history rather than starting the day you sign up, which is the whole point, because a before-and-after read taken on a thin window is not a read at all.

I have no stake in how much you advertise. I do not take a cut of your sales, I am not paid by Amazon, and there is no version of this where you spending more makes me richer. I get a subscription, and I keep it for as long as your numbers are going the right way. That is the whole alignment, and it is worth more than any feature on this page.

Chat is only half of it

Every other way of doing this makes you choose. Amazon's plugin is a chat window and there is nothing to look at. The MCP services that have appeared this year are the same shape, an endpoint and a conversation. The dashboards you already know are the opposite: something to look at, built on their copy of your data, behind their login.

I wanted both, because they are genuinely different jobs. An AI is the right tool for a question you did not know you had, the sort that would otherwise mean opening four reports and a spreadsheet. It is the wrong tool for Monday morning. Nobody wants to type "how did last week go" and wait for it to think about it. You want to open something and see.

Amazon Shopify TikTok Shop YOUR WAREHOUSE your own Google Cloud account, your billing years of history, refreshed every night ASK IT Claude or ChatGPT, arriving with fifty-three tools for the question you didn't know you had LOOK AT IT your own dashboard, built on the same rows for the Monday morning read
One warehouse, two ways in. Because they read the same rows there is no second version of the truth to reconcile, which is the usual reason two tools disagree about what you earned last month.

The dashboard is the part people underestimate until they have it. What sits on it:

It can also change things on Amazon: price, timed sales, listing fields, images, variation families, bids, negative keywords. Every change is validated against Amazon first, shown to you exactly as it will land, and applied only once you say yes. Then it is logged. The rail that can start new spend is the strictest of the lot, because that is the one where a mistake keeps costing you after you have stopped looking at it.

And the warehouse is yours. It is not a login to my system. If you stop paying me the data stays where it is, in your cloud, still queryable, and I will hand over the code and the runbook if you would rather run it on yourself. There is no access to lose, because you were never renting access in the first place.

The kind of thing it turns up

Four findings off my own account. None of them are visible in Seller Central, and none of them would ever have come from a tool that works for Amazon.

Learned expensively, on my own stock

Where you put a keyword decides whether it indexes. We tested the same terms in different places on the listing. A term in the backend search terms field indexed between eighty-nine and ninety-six percent of the time. The same term sitting in a bullet, sixty. Not on the listing at all, zero. So if there is a term you need to rank for, it belongs in the backend field, and putting it in a bullet instead is a coin flip you lose four times in ten.

Shallow and deep discounts go in opposite directions. Measured across the whole range, cuts in the five to fifteen percent band raised contribution per day. Past about nineteen percent they destroyed it. The difference between a promotion that works and one that just gives money away is a few percentage points, and you only find where your own line sits if you have the history to measure against.

Your own sale price caps the deal Amazon will approve. Amazon prices a deal off your trailing thirty-day low and wants it roughly five percent under. So a discount you run in October can quietly disqualify the Black Friday deal you were counting on, and nobody tells you until it comes back rejected.

Once a product sells organically, price buys rank, not volume. At that point the usual test, whether the P&L can afford to lose some percentage of units, is the wrong test, and applying it anyway is how a good listing gets talked into a slide that then takes months to climb back out of.

Who is writing this

I am Menachem Mendel (Mark) Saunders. I run the analytics for a supplement brand selling across six Amazon marketplaces and direct to consumer, and almost everything in S9 exists because I needed it, with a good deal of it built in the week after something went wrong and cost us money. I am based in the UK. The system runs on a US account without any of that mattering, and plenty of what I have had to solve, fee tiers, reimbursements, deal pricing, ad waste, is the same job in either market.

What it costs

S9 with the tools Your warehouse, nightly Amazon ingest, years of history backfilled, around fifty-three tools your AI can call, and the guarded write rails
$79 /mo
S9 with the dashboard All of the above, plus everything in the list further up this page
$129 /mo
Each further marketplace One Amazon marketplace is included. Add whichever others you sell in.
+$79 /mo

Both start with a free trial. There is no setup fee, no per-seat charge and no cut of your sales.

And it does not go up when you do. Most of this industry prices you by your revenue band, or meters you by rows and tokens, so the better your year the bigger the invoice. There is nothing here to meter. You bring your own Google Cloud account, which takes a few minutes to open and which Google bills you for directly at a few dollars a month, and you bring your own AI subscription. I am not sitting in the middle of either one taking a margin on it. It is your data and your usage, and the number above does not move when the business gets bigger. No tier traps, and no penalty for a good year.

Cancel whenever you like and the ingest simply stops, leaving the warehouse and everything collected up to that point sitting in your own account.

If that sounds too cheap to be any good

I know how $79 a month reads after everything above it. So, plainly: AI has made building this sort of thing far cheaper than it was two years ago. What would once have been a developer and half a year is now a run of evenings, and the running cost stays small because the two expensive parts, the cloud and the AI itself, you are buying directly rather than through me.

You could probably build it yourself. I am not going to pretend otherwise. I would only say that I would not spend those evenings to save $79 a month, and I doubt you would either.

In any case the code was never the hard part. The hard part is knowing which questions are worth putting to Amazon's data, and which of Amazon's own reports will quietly mislead you if you take them at face value. I have that because I sell, not because I build.

Try it before you pay for it

The trial is the real thing rather than a cut-down version of it: your warehouse built, your history loaded, the tools installed and your AI connected to all of it. The first thing it tends to turn up is money Amazon already owes you, usually inbound shipments where they received fewer units than you sent and refunds where the item never came back. On my own account that came to five figures, and most of it had been sitting there well over a year.

It hasn't just helped the business. It's helping every day.

Aryeh Leib Lopian, London

If you would rather hear that from him than from me, say so and I will put you in touch.

Email me and say which marketplace you sell in.
hello@s9search.com