An MCP server is how an AI app like Claude or ChatGPT connects to other software. Amazon and Shopify brands have three kinds to choose from. A platform's own server sees that one platform deeply. Data-pipe connectors pass reports from many sources, often with a free plan or a cheap trial. A joined, governed model defines numbers once across channels. Many brands use more than one.
An MCP server is the standard plug that lets an AI app like Claude or ChatGPT reach other software, such as your Shopify store or your Amazon account. For a brand on Amazon and Shopify, the best MCP server depends on the job: the platform’s own server for one platform, a data pipe for quick reports, and a joined model for numbers that must match.
Most “best MCP server” lists rank tools as if they did the same job. Below we sort them into three groups, say where each stops, and give a test you can run on any of them. Every claim about another company comes from its own site, as of September 2026. These products change often, so check before you buy.
What an MCP server is, in plain words
MCP stands for Model Context Protocol. It is an open standard for connecting AI apps to outside systems. The project’s own site says to think of it “like a USB-C port for AI applications” (modelcontextprotocol.io). Claude, ChatGPT and many coding tools support it.
The client is the AI app you type into. The server sits in front of some software and offers the AI a list of tools. A tool is a named action, such as “get orders between two dates.” The AI picks a tool, calls it, reads what comes back and writes an answer. Custom connectors you add in Claude are MCP servers (Claude docs).
MCP moves data. It does not decide what the data means. Two servers can both return “revenue” for last month and disagree by 10%, while both follow the protocol perfectly. So the useful question is not “does it support MCP?” It is “what does the AI get handed, and who decided what it means?”
Group 1: the platforms’ own MCP servers
These come from the platform itself and talk straight to its live systems.
Shopify
As of September 2026, Shopify lets merchants connect a store to Claude, ChatGPT or Perplexity (Shopify Help Center). Shopify’s page for the Claude connector says it can read orders, analytics, customer records and inventory. It can also edit products, collections, inventory and percentage-off discount codes, and it asks before it makes a change. It cannot refund or cancel orders, or change settings such as your plan, payments or taxes. Shopify says there is no extra Shopify fee for it, and you pay Anthropic for your Claude plan (Shopify connector for Claude).
Shopify also runs servers built for other jobs:
- Dev MCP gives coding tools Shopify’s developer docs and API schemas. An API is the way one piece of software talks to another. It runs on your computer (Shopify Dev MCP).
- Shopping servers for catalog search, carts, checkout and order status let AI agents buy on a shopper’s behalf (Shopify agent docs). They serve shoppers, not your finance team.
Amazon
As of September 2026, Amazon’s options are built for developers and ad partners, not for a founder with a chat window:
- The Amazon Ads MCP Server was announced in open beta on February 2, 2026. It is for Amazon Ads partners with active API credentials. It covers campaign changes, reporting, account settings and billing (Amazon Ads).
- Example MCP servers for the Selling Partner API (SP-API, Amazon’s seller API) sit in Amazon’s GitHub. Amazon calls them educational examples for developer work. Live calls to your account need your own SP-API keys (Amazon on GitHub).
We found no ready-made Seller Central connector from Amazon like Shopify’s.
Good for, and where it stops
Good for: live questions about one platform. The platform’s own server reaches its whole set of objects, including ones outside tools may not copy. A brand-new question still gets an answer, because the server asks the platform directly. Shopify’s connector can also make changes, and Shopify charges no extra fee for it.
Where it stops: one platform per server. Shopify’s connector cannot see your Amazon fees. It answers with Shopify’s definitions, which are not Amazon’s, and nothing reconciles the two. On Amazon, both official options are built for people who hold API access.
Group 2: data-pipe connectors
A data pipe is a service that pulls reports from Amazon, Shopify and ad platforms, then hands them to Claude through its own MCP server. Some work as a URL you paste into Claude. Here are four, each checked on its own site in September 2026:
| Tool | Sources its site lists | Can it change your accounts? | Entry offer on its site |
|---|---|---|---|
| Windsor.ai | Amazon Seller Central with 350+ sources | Yes: a listing’s title, description, bullets or price, if your Amazon app has the listing role | Free plan |
| Porter Metrics | 15+ sources, Amazon Seller among them | No, read-only | Free plan and 14-day trial |
| Adzviser | 40+ sources, Amazon Seller Central among them | No, read-only | $0.99 for a 14-day trial |
| DataDoe | Amazon only: Seller Central, Vendor Central and Ads, in one schema | Yes: prices, listings and ads, with an approval step first | $97 a month, 14-day free trial |
In our view, DataDoe sits partway toward group 3, for Amazon alone, because it cleans Amazon data into one set of tables.
Good for: speed and price. The sites describe setup in minutes, and you can put Amazon, Shopify and ad numbers in one chat without a developer.
Where it stops: a pipe passes along the fields each platform reports. Unless the vendor says otherwise, the joining and defining happen inside the chat, each time you ask. Amazon’s sales data alone offers both ordered product sales and shipped product sales. If one chat picks one and the next picks the other, you get two unlabeled “revenue” numbers for the same month.
Before you buy, ask each vendor three things. Which revenue definition does it use for Amazon? Does it store history, or ask the platform live each time? Does it label a number that is still changing?
Group 3: a joined, governed model
The third kind does the joining and defining once, before anyone asks. It keeps one model of the brand across every channel. That model says which Amazon listings and Shopify variants make up each product family, what counts as revenue on each channel, and which costs belong where. Every answer shares those written-down definitions.
Good for: questions that cross channels and must come out the same way every time. “Net revenue by channel last month.” “Margin on our best product family.” Numbers a board or a lender will see.
Where it stops: it costs more than a pipe and takes longer to set up, because someone has to write the definitions down. It answers what has been modeled. A question about a platform object nobody modeled may need new work, where the platform’s own server would answer it live. For changing a listing or a campaign, use the platform’s own tools.
This is the group we build in. Synthesis is ontology software and an AI harness for consumer brands. It joins Amazon, Shopify, ad platforms and more into one warehouse per brand. Every number carries its revenue basis, its period and its caveats. When two figures for the same metric and period differ by more than 5%, the answer is flagged instead of one being quietly picked. Our longer comparison with the platform servers, including the rows we lose, is at Synthesis vs. the vendor MCPs.
The three groups side by side
This table is our judgment, drawn from the vendor pages above.
| Platform’s own server | Data-pipe connector | Joined, governed model | |
|---|---|---|---|
| Sees | One platform, fully and live | Many sources, as report fields | Every connected source, joined |
| Cost | No extra Shopify fee; you pay for your AI app | Free plans and trials exist | Usually the highest of the three |
| Setup | Minutes on Shopify; developer work on Amazon | Minutes | Longest: sources, then definitions |
| Depth inside one platform | Deepest of the three | The fields each report exposes | What has been modeled |
| A question nobody planned for | Strongest: asks the platform live | Fine, if the field exists | May need new work |
| Makes changes | Yes, within its limits | Some do | Use the platform’s tools |
| Who defines “revenue” | The platform, its own way | Often the AI, each time you ask | Written once, shared by every answer |
| Joins Amazon to Shopify | No | Often left to the AI in the chat | Yes, before you ask |
| Labels a number still changing | Depends on the platform | Check each tool | Yes, with basis, period and caveats |
| Pick it for | One-platform facts and edits | Fast, cheap reads across sources | Numbers that must match every time |
Where it goes wrong
These traps show up whichever group you pick.
The same word, different money
Amazon has at least three honest answers to “what did we sell last month?” Ordered product sales counts orders when they are placed. Shipped product sales counts items when they ship. The settlement report shows what Amazon actually paid, after fees and refunds. Shopify adds gross sales and net sales. An AI that adds Amazon ordered sales to Shopify net sales has added two different things. See Amazon settlement report vs the orders report.
Fresh numbers read as final
Amazon keeps crediting sales to an ad click for days after the click. So last week’s ad sales will be higher next week. A raw week-over-week comparison shows a drop that is not real. See why Amazon ad sales numbers are not final. The open month has the same problem: half a month next to a full month looks like a fall.
A customer count built from missing data
Amazon treats buyer details as restricted personal data. An app needs a special token to read them (SP-API docs). So Amazon customer counts are built from partial data, and an exact figure with no caveat hides a gap.
Products that do not line up
An Amazon listing has an ASIN (Amazon’s product code). A Shopify variant has a SKU (the seller’s own stock code). A 3-pack of candles on Amazon and a single candle on Shopify belong to the same product family, but neither platform knows that. Someone has to write that mapping down. See product hierarchy for consumer brands.
An empty answer read as zero
A connection that broke last Tuesday returns nothing. Nothing looks like zero, and zero looks like a bad week. Ask any tool what it shows when a source stops syncing.
How to test any MCP server in 20 minutes
Do not take accuracy on trust, including ours. Pick five questions you already know the answer to, and write your answers down first. Then connect the server, ask each question in a fresh chat, and compare.
First, a worked example of a wrong answer. Illustrative numbers only: say a travel mug brand closed August 2026 with these figures.
| Figure for August | Amount |
|---|---|
| Amazon ordered product sales | $210,000 |
| Amazon product sales in settlement, after promotions and refunds | $188,000 |
| Shopify gross sales | $92,000 |
| Shopify net sales, after discounts and returns | $81,000 |
Now ask “what was net revenue in August?” One tool adds ordered sales to gross sales and says $302,000. Another adds settlement sales to net sales and says $269,000. Both are sums of real numbers. They are $33,000 apart, about 11% of the larger one. A good answer names the basis it used for each channel. A better one shows both and says why they differ.
The five questions:
| Question | Check it against | A wrong answer looks like |
|---|---|---|
| 1. Last closed month’s net revenue, per channel | Your month-end close | One total with no basis named. Amazon ordered sales added to Shopify net sales. The current month treated as closed. |
| 2. Best-selling product family across channels | Your own product list | A single listing ranked as a family. Amazon and Shopify names left unmatched. A ranking from one channel only. |
| 3. Amazon ad sales for one week, asked twice, a week apart | Amazon’s ad console on both dates | The number changes, and neither answer warned that it could. |
| 4. How many customers bought on Amazon last month | Nothing exact exists, which is the point | One exact number with no caveat. A figure equal to the order count. A sharp drop in the latest weeks. |
| 5. Margin on one product | Your cost sheet and a closed settlement month | No product cost asked for or cited. Margin before Amazon fees. Fees taken from a month still settling. For the full walk, see contribution margin. |
An answer that says “I can’t tell that from this data, and here is why” passes. A confident number with no label fails, even when it lands close. Close by luck will not stay close next month.
How to choose
- List the ten questions your team asks most. Mark each one “one platform” or “crosses channels.”
- Use the platform’s own server for one-platform questions and for changes. It gives the fastest, deepest answer about that platform.
- Add a data pipe for quick reads across sources. Check its revenue definition first, and have a person review numbers before they go anywhere important.
- Add a joined model when numbers must match every time. Board decks, lender reports and budget calls are the usual cases.
- Run the 20-minute test on your shortlist. Same five questions, a fresh chat each time.
- Expect to run more than one. The kinds do different jobs and can sit in the same chat.
Questions people ask
What is the best MCP server for ecommerce?
There is no single best one, because the three kinds do different jobs. Use a platform’s own server, such as Shopify’s connector for Claude, for questions about that platform and for making changes. Use a data pipe for quick reads across sources, and a joined model when cross-channel numbers must match every time.
Does Amazon have an official Seller Central MCP server?
As of September 2026, Amazon posts example MCP servers for Selling Partner API developers on GitHub. It also runs an Amazon Ads MCP Server in open beta for Amazon Ads partners with active API credentials. We did not find a ready-made Seller Central connector for merchants that works like Shopify’s.
Does Shopify have an official MCP server?
Yes. As of September 2026, Shopify lets merchants connect a store to Claude, ChatGPT or Perplexity. Its connector for Claude can read orders, analytics and customers, and edit products, inventory and discount codes. Shopify also runs a Dev MCP server for developers and shopping servers for AI buying agents.
How do I connect Amazon Seller Central to Claude?
There are three routes. A developer can run Amazon’s example Selling Partner API server with your own API keys. A data-pipe service can connect your seller account and give you a URL to paste into Claude. Or a joined model can connect Amazon with your other channels and serve the combined data.
Can an MCP connector change my store or listings?
Some can. Shopify’s connector for Claude can edit products, inventory and discount codes, and some data pipes can edit Amazon listings. Read the permissions screen before you approve, and pick a read-only tool if you only want answers.
Why do two AI tools give me different revenue numbers?
Usually because they used different definitions or dates, not because one is broken. Amazon alone has ordered sales, shipped sales and settlement figures. Ask each tool which basis, period and marketplace it used.