What Is MCP? A Plain-English Guide for E-commerce Merchants

If you run an online store, you have probably seen "MCP" appear in platform announcements and agency pitches. This guide explains what it is, what it does for a merchant today, and where it falls short. No coding knowledge is needed.

What is MCP in plain English?

MCP stands for Model Context Protocol. It is an open standard that Anthropic introduced in November 2024. It defines one common way for an AI assistant to connect to outside tools and data, such as your store catalog, your ad reports or your email platform.

A rough analogy: before USB, every device needed its own cable. MCP aims to be the shared plug between AI assistants and business software. Without it, each AI app needs a custom integration with each tool. With it, a tool publishes one MCP connection and any compatible assistant can use it.

The official specification and documentation live at modelcontextprotocol.io. It is not owned by a single product, and it has since been adopted across the industry. Microsoft, for example, added MCP support to Copilot Studio in March 2025.

How does MCP work?

Three terms cover most of it.

• MCP client: the AI app you talk to, such as Claude. It connects to servers on your behalf.
• MCP server: a small service in front of a tool or data source. It describes what the assistant is allowed to do there.
• Tools, resources and prompts: what a server exposes. Tools are actions (search products, pull a report). Resources are data the assistant can read. Prompts are reusable instruction templates.

Servers can run locally on a computer (called stdio) or remotely over the web (HTTP). Remote servers are the usual choice for hosted platforms such as a store or an ad account.

In practice you ask the assistant a question in normal language. It decides which tool to call, the server fetches the data, and the assistant answers using the result. The server, not the assistant, decides what is possible.

Why does MCP matter for an e-commerce store?

Most store work involves moving information between systems: orders, ad spend, email performance, customer questions. Today that often means exports, spreadsheets and dashboards. MCP lets an assistant query several of those systems in one conversation.

There is also a second, newer angle. Shoppers are starting to use AI assistants to find and compare products. MCP is one way those assistants can read a store's catalog and policies directly, rather than guessing from scraped web pages. So MCP matters both for how you run your business internally and for how AI tools may see your store externally.

Which e-commerce tools offer MCP today?

Support varies a lot. This list reflects what is documented, and details change quickly, so check each vendor's own docs before relying on it.

• Shopify: the most mature platform support. It offers Dev MCP for developers and Storefront MCP endpoints for stores. Shopify's docs describe servers for a single store's catalog, cart and policies, and a separate catalog server that searches across the Shopify ecosystem.
• Google: publishes open-source MCP servers for Google Ads (read-only reporting) and Google Analytics.
• Klaviyo, Stripe, HubSpot and Zapier: each offers an MCP server.
• WooCommerce: has native MCP support, but its developer documentation labels it a developer preview, meaning details may change.

Third-party articles disagree on some Shopify specifics, such as which endpoints are enabled by default for every store, when that began, and which pieces (such as checkout) are still in preview. Treat those claims as unverified until you confirm them on shopify.dev.

How is MCP different from an API or a plugin?

	Plugin or app	API	MCP
Built for	Merchants using a dashboard	Developers writing code	AI assistants
Setup	Install and configure	Custom development	Connect a server to an AI app
Flexibility	Fixed features	Anything the API allows	Whatever the server exposes, asked in plain language
Typical user	Store owner	Developer	Merchant or agency using an AI assistant

MCP does not replace APIs. Most MCP servers sit on top of an existing API and package it so an assistant can use it safely. It also does not replace your platform admin. Shopify's own servers, for instance, focus on reading and shopping tasks, and they do not edit themes. Anything beyond that needs a different server, often from a third party, with its own login.

What is the relationship between MCP and UCP?

You may also hear about UCP, the Universal Commerce Protocol. Google announced it in January 2026 as an open standard for agentic commerce, meaning AI agents that help people discover, buy and get support for products. It was co-developed with Shopify, Etsy, Wayfair, Target and Walmart.

A simple way to separate them (a framing, not an official definition): UCP describes what a store can do and how a shopping agent should ask. MCP is one of the ways the agent actually connects. Shopify's catalog servers are described in its docs as an MCP binding of UCP. Payment authorization is handled by a separate protocol, AP2.

For most merchants, the takeaway is that these standards are meant to work together. You will mostly meet them as settings and features inside your platform, not as something you build.

What are the limits and risks of MCP?

MCP is useful, but it is not magic. Be clear about these points before you connect anything.

• It exposes access, not judgment. An assistant can misread data or draw a wrong conclusion. Check numbers that drive spending decisions.
• Permissions matter. A server that can only read reports is low risk. One that can change prices, refund orders or edit products is not. Prefer read-only access to start, and give the narrowest permissions that do the job.
• Untrusted content can be a problem. If an assistant reads text written by outsiders (reviews, support tickets, product descriptions), that text could contain instructions meant to steer it. This is called prompt injection. Do not let an assistant take high-impact actions without a human approval step.
• Third-party servers vary in quality. Anyone can publish an MCP server. Prefer servers from the platform vendor, and review who maintains any other server and what data it can see.
• Features are still moving. The spec has been revised several times, and platform support, such as WooCommerce's developer preview, is still maturing. Expect changes.
• Data quality still decides results. An assistant querying messy tracking or incomplete cost data will give you confident answers built on weak inputs.

This guide does not cover MCP security in depth. Read each vendor's security and authorization documentation before connecting a live store.

Do merchants need to understand the MCP spec version?

No. The specification is versioned by date, and the maintainers have released several versions since 2024. Platforms, apps and AI tools handle version negotiation between clients and servers. Older and newer versions can coexist for some time. If you are a merchant or agency lead rather than a developer, you can skip this detail.

How should a merchant get started with MCP?

A cautious path:

1. Pick one low-risk question. For example, "which products drove the most ad spend last month" or "summarize our shipping and returns policy gaps."
2. Use a vendor-provided, read-only server first. Google Ads reporting and Google Analytics are natural examples because they are read-only for reporting.
3. Connect it to an AI app you already use. Follow the vendor's setup instructions rather than a random tutorial.
4. Verify the answers. Compare the assistant's output with the native dashboard a few times before trusting it.
5. Add write access only with approval steps. If a task changes live data, keep a human confirmation in the loop.
6. Document what is connected. Keep a list of servers, owners and permissions, especially if an agency set them up.

If you work with an agency, ask which servers they use, who maintains them, and whether access is read-only. Those three questions reveal a lot.

Is MCP worth paying attention to now?

Yes, in a measured way. The protocol is open, major platforms and vendors support it, and the practical use cases (reporting, catalog queries, customer service lookups) are real. But much of the agentic shopping side is new, with preview features and conflicting third-party reports. Learn the basics, test on read-only data, and avoid building core processes on features the vendor still calls a preview.