You write a request
You describe what you want in your own words, the way you would brief a colleague. You do not name any tools.
How the AI tool you already use gets to work on your store, explained without the developer vocabulary
Published
MCP, the Model Context Protocol, is an open standard that lets an AI tool use another system's functions.
For a store owner that means one thing: The AI tool you already write to every day can look at your store's data and change how the store behaves, because the systems behind the store have given it a set of actions it is allowed to take.
Most explanations of MCP are written for developers. This one is written for the person who runs the store and wants to know what the connection does, what it can touch, and what stays in their hands.
An AI tool on its own knows a great deal about the world and nothing about your store. It has never seen your catalog, your search log or last month's newsletter results. To be useful for the store, it needs a way in.
The Model Context Protocol is the agreed way of building that door. A company that holds your data runs an MCP server, which is a list of actions with a description of each. Your AI tool connects to the server, reads the list, and from then on can use those actions when a request of yours calls for them.
Because MCP is an open standard, one connection works across tools. Hello Retail documents setup for Claude, ChatGPT and OpenAI Codex.
You describe what you want in your own words, the way you would brief a colleague. You do not name any tools.
It reads the descriptions of what is available and decides which actions answer the request. A question about empty searches might use one action to fetch the list and another to check whether you sell a match.
Each action runs on the system that holds the data, using your own sign-in. No password or API key is pasted into the AI tool, and it gets only the result back.
The tool tells you what it found. If the next step would change something, it says what and waits for your yes.
They fetch something and change nothing: Which searches found nothing, how a recommendation box performs, when the product feed last ran. You can ask as often as you like.
They alter how the store behaves: Add a synonym, put products first on a sale page, set up a newsletter campaign. The AI tool asks before it uses one.
In the documentation these are called read tools and write tools. The difference between them is the most useful thing to understand about the whole subject, and our guide to what your AI tool can change in your store covers it in full.
The Hello Retail connection gives an AI tool actions for Search, Recommendations, Pages, Newsletter Content, Product Agents, Retail Media and Product Intelligence, and for product data and feeds. The table is read from the live server, so it follows new tools as they ship.
| Area | Tools that look | Tools that change |
|---|---|---|
| Search | 24 | 28 |
| Recommendations | 13 | 10 |
| Pages | 12 | 13 |
| Newsletter Content | 16 | 8 |
| Product Agents | 3 | 2 |
| Retail Media | 4 | 3 |
| Product Intelligence | 1 | 0 |
| Product data and feeds | 13 | 5 |
| Orders | 1 | 0 |
| Operations | 7 | 2 |
For what those actions add up to in practice, see what you can build, product by product.
The Hello Retail MCP page shows the connection at work.
Tool names, request formats and the protocol itself can stay out of sight. They matter to the people who build the connection, and the reason the standard exists is so that the people who use it do not have to think about them.
Model Context Protocol. It is an open standard, first published by Anthropic in November 2024, that describes how an AI tool connects to an outside system and uses that system's functions as tools.
It is the connection that lets the AI tool you already use look at your store's data and make changes to your store's search, recommendations, pages and email content when you ask it to.
No. MCP is a standard, like a plug shape. A company that supports it runs what is called an MCP server, and you connect your AI tool to that server. At Hello Retail you sign in with your existing account, and the tool gets the same access that account already has.
No. You need to know three things: Which store the tool is connected to, whether a request only looks or also changes something, and where a change goes after you approve it. The rest is handled between the tool and the server.
The safeguards to look for are a personal sign-in, a confirmation step before any change, drafts that a person publishes, and a log of changes. The Hello Retail connection has all four. A few kinds of change take effect when saved, and they are listed in our guide to what your AI tool can change.
An API is what developers write code against. MCP packages the same functions so that an AI tool can use them from a request written in ordinary language. Our guide to MCP and APIs for ecommerce compares the two.
The setup takes a server address, a sign-in and your Company ID. Our guide to connecting your AI tool to your store covers the first session.
Start building with our MCP