An Amazon Vendor Central MCP is a connection that lets an AI client such as Claude or ChatGPT read your first-party vendor data, and in some cases act on it, through the Model Context Protocol. You ask about purchase orders, fill rate or Net PPM in plain English, and the client fetches the answer from your account. DataDoe runs one of these connections, described on our Vendor Central MCP page, but it is not the only kind.
This guide covers what a Vendor Central MCP is, why Amazon's own Selling Partner plugin does not reach vendor accounts, the three types of option a 1P vendor has and how to judge them, what an AI client can read and change, how to connect Claude and ChatGPT step by step, example questions, and the limits. Amazon's pages and the other providers' public pages were read on 11 November 2026.
MCP, the Model Context Protocol, is an open standard for giving an AI client tools. A server publishes a list of things it can do, and the client decides which one to call while it answers you. A Vendor Central MCP is such a server sitting in front of vendor data.
Underneath, every one of them uses the same source: Amazon's Selling Partner API, where the vendor side has its own set of APIs and reports. None of them reads the Vendor Central screens. So a server can only return what the vendor APIs return, and what you can ask depends on which of them the provider chose to cover.
Not one that we could find for vendors. Amazon announced the Amazon Selling Partner plugin on 23 September 2026. It launched in Amazon Quick and is in beta in Claude, for sellers in Amazon's U.S. stores, with international expansion to follow. The connector listing in Claude's directory tells you to connect an Amazon Seller Central account. Neither Amazon's announcement nor that listing mentions vendors or Vendor Central.
Amazon Ads has a separate MCP server in open beta for partners with Ads API credentials. It covers advertising, not retail data.
A brand that sells both ways can use Amazon's plugin for the third-party side in the U.S. only. Our comparison of Vendor Central and Seller Central explains why the two sides do not share reports or fields.
A hosted data layer fetches your vendor data on a schedule, stores it in tables and puts an MCP endpoint in front of the tables. You authorize it in Vendor Central and paste one URL into your AI client. Stored data keeps history that Amazon no longer serves and can be joined to ads and to a seller account. Some layers also offer write access. The trade-off is that the data is as fresh as the last fetch. DataDoe is in this group.
Reporting connectors that send marketing data to spreadsheets have added MCP as one more destination. Porter's page describes a hosted, read-only server with 17 Vendor Central fields (ordered and shipped revenue, shipped COGS, units and customer returns, by day and by hour), a free plan and paid plans from $15 per account a month. Adzviser lists a Vendor Central connector with 7 metrics and 20 breakdowns that feeds Sheets, BI tools and AI clients from one subscription. They fit sales reporting next to other channels and are narrower on purchase orders, invoices and inventory.
The third kind exposes the vendor API operations themselves as tools, one tool per operation. Kuudo's README lists 40 tools across the Retail Procurement and Direct Fulfillment APIs, 16 of them write tools, running in a deployment in your own cloud. Community servers on GitHub follow the same pattern with your own SP-API credentials. This route gives the most direct control, but there is no stored history and each question becomes a series of live API calls.
Amazon documents the vendor side of the Selling Partner API in a few groups:
Two Amazon rules shape every answer. The periodic reports arrive with a lag: Amazon says daily data is due 72 hours after the day closes, so no server can show yesterday's final shipped revenue this morning. And some report windows are short, for example seven weeks of weekly data.
In DataDoe these arrive as Vendor Central tables: sales, traffic and inventory per ASIN and for the whole catalog at daily, weekly and monthly grain, the weekly demand forecast kept as a new snapshot each week, the three hourly tables, purchase orders and purchase order status, shipments, Direct Fulfillment orders, invoices and invoice lines, remittances and remittance lines, vendor listings, promotions and coupons. Purchase orders are checked every two hours. Most other tables refresh in the daily fetch that starts at 2am in the vendor's local time. What each report measures is the subject of our guide to Amazon Vendor Central reports.
Amazon's vendor APIs include write operations: accept or reject a purchase order, submit a shipment or a shipment confirmation, acknowledge a Direct Fulfillment order, submit an invoice. Whether an MCP exposes them is the provider's decision.
DataDoe exposes a defined set as Actions. For vendor accounts that set is:
Invoices are read-only in DataDoe: it does not submit them. Setting a price is a seller-side Action and is not available for vendors. Actions stay off until you enable them, most AI clients ask before a change, and every change is logged. A write also needs the account set to read and write access, and any Action can be sent as a dry run first.
Whichever tool you choose, start read-only. A purchase order acknowledgement is a commitment to Amazon, so let the AI prepare the list and have a person confirm it.
The steps below use DataDoe.
Custom connectors work on every Claude plan, and the Free plan allows one. In a Team or Enterprise workspace an Owner adds the connector under Organization settings first.
ChatGPT's menu labels differ between accounts, and on Plus or Pro plans write actions through a custom app may not be available, so test on your own account. If you would rather build against the data directly, the Amazon Vendor Central API page shows the same tables over REST.
Name the view and the period in every question.
We found none. On 11 November 2026 Amazon's Selling Partner plugin connected Seller Central accounts and was in beta for sellers in Amazon's U.S. stores. Vendors use third-party options that sit on the vendor side of the Selling Partner API.
Both. MCP is an open standard, so one server URL works in Claude, ChatGPT and other MCP clients. ChatGPT needs Developer mode for a custom app.
With a hosted option, usually not: you approve the provider's application in Vendor Central.
Technically yes, where the server exposes that operation. In DataDoe it is an Action that stays off until you enable it, and it can be run as a dry run first.
DataDoe is the Amazon data layer for AI agents. It keeps Vendor Central, Seller Central and Amazon Ads in 130+ tables behind one MCP endpoint, at $97 a month after a 14-day free trial. The Vendor Central page lists the vendor tables and Actions in full and is the place to begin: connect one marketplace, let the first load finish, and put the purchase order question from the list above to Claude.
.svg)
