โ† Back to Blogs

Amazon Strategy

12 min read

Which Amazon MCPs Store Client Context (And Which Don't)?

Onieque Edwards

Content Strategist /Blog Writer

You connect an AI assistant to a client's Amazon ad account on Monday. You explain the brand: the margin on the hero product, and why its exact-match campaign is allowed to run hot. The answers that afternoon are sharp.

On Thursday you open a new chat and ask a follow-up. The assistant pulls the same campaign data and answers as though it has never heard of the brand.

An Amazon MCP server is the connector that lets an assistant read your Amazon data, and client context is everything about the account that the data does not say. Most of these connectors fetch the data and keep none of the context. For most providers, the documentation does not say either way.

This article sorts the providers by what they actually say, explains why silence is not the same as a no, and shows what stays missing even when a tool remembers everything.

In this article

  • What Is an Amazon MCP Server, in Plain Terms?

  • What Does "Client Context" Mean on an Amazon Account?

  • Which Amazon MCP Servers Store Client Context?

  • Why Is "Not Documented" Different From "Does Not Store"?

  • What Happens When an Amazon MCP Starts Every Session From Zero?

  • Is Stored Client Context a Privacy Risk?

  • Does an Amazon MCP That Remembers Replace Strategy?

  • So Which Amazon MCP Should You Pay For?

  • FAQ

What Is an Amazon MCP Server, in Plain Terms?

MCP stands for Model Context Protocol. It is a shared standard for plugging an AI assistant into another system, so the assistant can read data from it and, where you allow it, make changes.

An Amazon MCP server is the plug. One side talks to your assistant. The other side talks to Amazon's API, which is the doorway Amazon opens for approved software to read and change an account.

Amazon has its own for advertising. The Amazon Ads MCP Server went into open beta on 2 February 2026, according to Amazon, for advertisers and partners with active Amazon Ads API credentials. Amazon lists tools to create, update and delete campaigns, pull performance reports, and change account settings and billing. Third parties such as Marketplace Ad Pros, Seller Labs, Helium 10 and Adspirer offer their own connectors, some for ads, some for Seller Central data, some for both.

Here is the trap in the name. "Context" in Model Context Protocol means the live data handed to the assistant for the question in front of it, not memory of anything once the conversation ends.

A connector that only does that job is a stateless pass-through. It takes your question, fetches the numbers from Amazon, hands them back, and keeps nothing for next time.

What Does "Client Context" Mean on an Amazon Account?

Client context is what you would tell a new account manager on their first day that no report would tell them. On an Amazon account it usually covers:

  • The goal for this quarter, whether that is growth, profit or holding rank through a difficult stretch.

  • The margin on each product, and so how much ad cost a sale can carry.

  • Past tests, and what they showed.

  • Why a campaign is set up the way it is.

  • Stock plans, such as a reorder landing late or a product being phased out.

  • What has already been ruled out, and why.

None of that lives in the Ads API. A stateless connector starts every session from zero, so you either explain the account again or you get an answer built on the data alone.

The data alone goes wrong in a predictable way. It treats every number as though the number were the goal. A high ACoS (advertising cost of sale, the share of sales spent on ads) on a launch product looks like waste. A campaign paused after a failed test looks like a missed opportunity.

Which Amazon MCP Servers Store Client Context?

I sorted each provider by what it says publicly about keeping anything between sessions. Where a provider says nothing, the label is "not documented".

Provider

What it says

Label

Amazon Seller Assistant, Selling Partner plugin for Claude and Amazon Quick

Amazon lists "persistent memory of pricing patterns, inventory cycles, and growth goals"

Says it keeps context

Adspirer

Its FAQ describes a brand context file holding budget limits, checked before any action that changes spend, and strategy that carries between sessions. It also says it does not permanently store campaign performance data, ad copy or targeting settings, and keeps no conversation logs

Says it keeps context

Marketplace Ad Pros

An audit trail of every change, 90 days of history pulled at connection with no expiry, and an Experiments feature that keeps a record of past tests per brand. No feature for free-form goals or notes is described

Stores records

Seller Labs

Its Data Hub stores data from every linked marketplace, with no stated limit on history. Chat memory is not described

Stores data

Kuudo and the open-source KuudoAI server

Downloaded reports are saved to the server's disk, and "skills" define tasks. Client notes are not documented

Stores files

Two Minute Reports

Says it does not store advertising data on its servers. Memory between sessions is not documented

Says it does not store ad data

Amazon Ads MCP Server (official)

No memory feature in the material available

Not documented

Intentwise

Teaches the assistant terms like ACoS and TACoS. Persistent memory is not described

Not documented

Helium 10, Pacvue, Nova, DataDoe, Jungle Scout, Teikametrics, Ad Badger, Atom11, Agent Central, Sellerise, SellerSonar

Nothing found on memory or saved context

Not documented

That is 19 providers. Two say they keep context about goals or strategy. Three store data, records or files. One says it does not store ad data. Thirteen say nothing I could find.

Two notes on the table. First, Amazon does not call its Selling Partner plugin an MCP server, and it is in beta in Claude for US stores only. It is in the table because sellers meet it in the same place as the others. That its memory carries over into Claude was reported by The Next Web; Amazon's own page does not say it in those words.

Second, for several names in the last row, the only detailed sources are comparison guides written by vendors that sell competing tools. Treat that row as a starting point for your own questions.

Your chat tool may have memory of its own, such as files saved in a Claude Project. That belongs to the assistant, not to any Amazon MCP.

Why Is "Not Documented" Different From "Does Not Store"?

Silence can mean a provider keeps nothing. It can also mean it keeps something and has not written about it. From the outside you cannot tell which.

Each wrong assumption costs something different. Assume a tool remembers, and you rely on context that is not there. Assume it stores nothing, and you may type a client's margins into a system that keeps them.

Even the clear statements are the providers' own. Two Minute Reports is the only one in the table that states plainly it does not store ad data, and none of the retention claims here have been independently audited in anything I read. Read "not documented" as a question to put to the provider in writing, never as an answer.

What Happens When an Amazon MCP Starts Every Session From Zero?

Here is the same question asked twice. The brand and every detail below are invented, as a hypothetical illustration. They exist to show what changes, nothing more.

The question: "The main campaign on our hero product runs out of budget by mid-afternoon. Should we raise the budget?"

Session one, stateless connector. The tool pulls spend, sales, ACoS and the budget cap. The campaign caps out early, it is converting, and its ACoS sits above the account average. The answer is sensible on the data: raise the budget a little and trim the keywords pushing ACoS up.

Session two, a tool with stored context. The context file says the goal this quarter is to hold rank on the hero product. It says a broad-match test in the spring wasted spend and was stopped. It says a reorder is landing late, so stock will be thin for the next few weeks. And it says the expensive-looking keywords brought in the most first-time buyers last year.

Now the answer changes. Do not raise the budget into a stock gap. Keep the keywords that look expensive. Do not bring broad match back.

The stock warning only exists because a person wrote it into the file, since Amazon's Ads MCP covers advertising data and inventory sits elsewhere. If that note is three weeks old and the reorder arrived early, the tool gives a confident answer built on something no longer true.

And neither session asked whether the campaign caps out because a rival ran out of stock and its shoppers are landing on you. That changes the budget decision again, and an ads-only connector cannot see it.

The worked example is invented. The pattern in the chart below is not: of the 19 providers named in this article, 13 document nothing about memory.

If you are weighing a provider right now, the questions to put to them in writing are set out in Questions to Ask an Amazon MCP Provider Before You Sign Up.

Is Stored Client Context a Privacy Risk?

Stored context is client data, and it is the sensitive kind. Margins, stock plans and growth goals are exactly what a competitor would want to know. Once a provider stores them, they live on somebody's servers.

The statements about where that data goes do not always line up. Amazon says seller data in its plugin stays within Amazon's infrastructure. SellerSonar, quoting Amazon's help page, reports that once the data enters Claude it is processed on systems outside Amazon's control. I have not checked that against Amazon's help page, and both can be true at different stages.

Vendor promises, such as Marketplace Ad Pros saying it does not train models on your data, are not independently audited in anything I read.

Before any client's context goes into a tool that keeps it, get written answers to these:

  • Where does the stored context live, and in which country?

  • Who at the provider can read it?

  • Is each client's context kept apart from every other client's?

  • What happens to it when you disconnect or cancel?

  • Is any of it used to train AI models?

If you run accounts for other brands, read your own client agreements too. Each client's notes are that client's data, and an agreement that lets you hold them says nothing on its own about a third party holding them.

Does an Amazon MCP That Remembers Replace Strategy?

Take the best version in the table: a tool that remembers goals, budget limits and past tests. It holds the data and the notes. It does not hold the judgment.

The judgment is reading the ad numbers against the listing, the reviews, the stock position, the price and the rank, and deciding where the money goes.

Say ACoS climbs on a product. The cause could be rising cost per click. It could be conversion falling because a new one-star review pulled the rating down. A rival may have cut its price, or organic rank may have slipped so the ads are carrying more of the sales. Those causes call for different moves, and some call for opposite ones. An ads-only connector sees one of them.

Adbrew, which sells PPC software, makes the same point: an ads server carries out requests without judging the strategy behind them, and cannot see inventory, margin, the Buy Box or organic rank. Even Amazon's launch page concedes that a connection alone does not guarantee good outcomes.

That connected read is the work I do. Jungle Pundits is one operator, and the job is holding those numbers against each other until the flags that move together point at a single cause. One flag is noise. Correlated flags are a finding. A connector can fetch every one of those numbers. Deciding which one explains the rest, and what the next dollar should buy, is still a person's call.

So Which Amazon MCP Should You Pay For?

Start with what you are buying.

If you want fast answers from live data and you will hold the context yourself, a pass-through connector does the job. It also stores less of your clients' information, which leaves fewer places for it to leak.

If you are paying for memory, pay only for memory that is written down: what it keeps, where, for how long, and who can read it. "Not documented" is not a feature, whatever the sales page implies.

Neither one makes the budget call. So before your next renewal, put two numbers side by side: what the subscription costs you a month, and what one wrong budget decision on your hero product costs during a stock gap. Then spend on whichever of the two you can least afford to get wrong.

Your data is connected. Is anyone reading it together?

If your ad numbers come back in seconds but nobody is checking them against stock, reviews, price and rank, the fast answer can still be the wrong one. The Expedition Audit is that read, done on your account, ending in a plain answer about where your ad money should go next. Book your Expedition Audit.

FAQ

Does the Amazon Ads MCP Server remember past conversations?

No memory feature is described in the material available. That is "not documented", not a confirmed no, so ask directly.

Does Amazon's Seller Assistant plugin for Claude have memory?

Amazon says Seller Assistant keeps memory of pricing patterns, inventory cycles and growth goals, and The Next Web reported that it carries over to Claude. The plugin is in beta in Claude for US stores.

Which Amazon MCP providers say they store context?

Of the 19 providers in this article, Amazon's Seller Assistant plugin and Adspirer say they keep context such as goals or strategy. Marketplace Ad Pros, Seller Labs and Kuudo store data, records or files. Two Minute Reports says it does not store ad data. The other 13 do not document it.

Is it safe to store client margins and goals with an MCP provider?

Only once you have written answers on where it lives, who can read it, and what happens when you disconnect. The retention claims here are the providers' own, not audited.

Can I add memory to an Amazon MCP that does not have it?

Your chat tool may keep its own, such as files in a Claude Project. You maintain that yourself, and a stale note gives a confident wrong answer.

Does an Amazon MCP with memory replace an Amazon agency?

It replaces the fetching. It does not decide which connected number explains the rest, or where the next dollar of ad spend should go.

Sources

Amazon's open beta date, eligibility and tool list for the Ads MCP Server come from Amazon's announcement Introducing the Amazon Ads MCP Server, 2 February 2026, which also carries its caution that a connection alone does not guarantee outcomes. The Seller Assistant memory, the data-stays-within-Amazon statement and the plugin's US beta status come from Amazon's Seller Assistant and Selling Partner plugin page (no visible date). That the memory carries over to Claude is from The Next Web, 23 September 2026. Adspirer's brand context file, and its storage and conversation-log statements are from the Adspirer FAQ. Marketplace Ad Pros' audit trail, history, Experiments feature and training statement are from marketplaceadpros.com. Seller Labs' Data Hub and history are from its MCP FAQ and pricing guide, updated 21 October 2026. Kuudo's skills are from kuudo.com, and report storage on disk from the KuudoAI amazon_ads_mcp repository. Two Minute Reports' statement on ad data is from its Amazon Ads integration page. Intentwise's ACoS and TACoS layer is from its AI Gateway page. Helium 10's MCP is described on its blog, 16 July 2026. The statement that no memory feature is described for any server is from Marketplace Ad Pros' Top Amazon MCP Servers guide, updated 4 November 2026, and SellerSonar's seller MCP comparison, updated 4 November 2026, is silent on it; both are written by vendors selling competing tools. SellerSonar's report of Amazon's help page on data processed in Claude is from its Selling Partner plugin guide, as of 16 October 2026. Adbrew's point on what an ads-only server can and cannot see is from its Amazon Ads MCP article. Status labels were compiled from these pages, read 9 November 2026.

Onieque Edwards

Content Strategist /Blog Writer

Onieque is the brain behind bold Amazon growth strategies and structured business execution. He enjoys turning scattered ideas into clear, actionable systems that actually drive results. When heโ€™s not building out growth plans or refining campaigns, youโ€™ll likely find him exploring new coffee spots or getting lost in ideas that connect strategy with creativity.