Amazon's Selling Partner plugin has been live in Claude for almost a month. Connecting it takes about a minute. Knowing what your AI can actually see once it is connected takes longer, so we did the slow part: we read every skill in Amazon's public toolkit, line by line, and put the same questions to an AI with and without the full picture.
Short version: the plugin is a good, free start for a single US seller who wants to fix listings and watch stock. It cannot see ads, Vendor Central, product costs or other marketplaces, and its numbers trail by about two days. If your AI can't answer a question, it usually can't see enough data to answer it.
New to the plugin? Start with our guide to the Amazon Selling Partner plugin for what it is and how to set it up. Jakob walks through the findings below in this video: Amazon Seller Assistant in Claude: 5 Things to Check.
The two answers pointed in different directions. The first AI could only rank products by what they sold. The second could weigh what each product actually earns after fees, ads and cost, and whether there was enough stock to sell more of it.
Neither AI made a mistake. Each worked from the data it had. That is the whole point of this review: before you compare AI tools, compare what they can see.
The plugin is good at the jobs it was built for. From the toolkit:
It is official, it is free during the beta, and it comes from the company that runs the marketplace. For a single US seller who mainly needs listing fixes and stock alerts, that is a lot for nothing.
These are the limits we found in the code, in Amazon's own words where possible.
The seller-analytics skill tells the AI that "data is delayed ~2 days, so the last two days would be empty", and to default to a window "ending at T-2". Ask how yesterday went and the answer is about the day before yesterday.
Every call is pinned to one seller and one marketplace: "Pin one entityId + one marketplaceId per call" and "don't blend accounts or marketplaces". A seller in the US, UK and Germany gets three separate answers that the AI has to add up itself. An agency gets one client at a time. And at launch, only US stores can connect at all.
No skill touches Amazon Ads. Ad data lives behind a separate Amazon Ads MCP server with its own credentials. With two connections the AI has to join ad spend to sales on its own, which costs time and tokens and is easy to get wrong.
The finance skill closes a month against payouts. It does not calculate profit by product, and nothing in the toolkit knows what your products cost you. Profit by SKU needs fees, refunds, ad spend and cost of goods in one place.
The plugin is for third-party sellers. Vendors have no first-party AI route yet.
The plugin reads Amazon's APIs on demand. There is no warehouse, no export and no SQL, so your history is whatever Amazon's APIs return on the day you ask.
Amazon lists Claude, in beta, and Amazon Quick. ChatGPT, Cursor, Gemini and agent frameworks are not on the list yet.
On the same example account we asked DataDoe: "Which of my advertised products will run out of FBA stock in the next 30 days, and how much am I still spending on ads for them per day?" It answered in one pass, because inventory and ad spend sit in the same schema. It listed the products that would run out within 30 days and their daily ad spend.
Then we asked it to cut bids on those products by 30% so the stock would last. DataDoe ran the change as a dry run first and showed exactly which bids would move, from what to what. Nothing reached Amazon until it was approved.
The plugin cannot answer the first half of that question, because it has no ad data, and cannot do the second half, because it has no ad actions.
This checklist works for any plugin, connector or MCP server, ours included. Paste it into your AI along with the tool's documentation.
Every customer also gets an account manager and a reply from a person in about 15 minutes. DataDoe is built by Deltologic, a software house that has worked with Amazon's APIs since 2020.
For a feature-by-feature view, see DataDoe vs Amazon Seller Assistant, or how to connect Amazon to Claude with DataDoe.
Try DataDoe free for 14 days. The plan is $97 a month after the trial. Start your trial.
Yes, for what it covers. It is official, free in beta, and strong on listing problems, stockouts and FBA inbound for US sellers. It does not cover ads, Vendor Central, product costs, other marketplaces or stored history.
Its analytics skill says data is delayed about two days and tells the AI to use windows ending two days ago.
No. Ads have their own Amazon Ads MCP server with separate credentials. DataDoe puts ads, sales, fees and inventory in one schema.
No. Its finance skill reconciles a completed month against payouts. It does not use product costs or ad spend, so it cannot give profit per SKU.
Not in one question. Each call is pinned to one seller account and one marketplace, and only US stores are supported at launch.
Yes. Connect both to Claude and the AI uses whichever one can answer.
DataDoe is not affiliated with or endorsed by Amazon or Anthropic. Plugin details come from Amazon's public selling-partner-agentic-toolkit repository as of 19 October 2026 and may change during the beta.
.svg)
