> For the complete documentation index, see [llms.txt](https://docs.elevateab.com/elevate-helpcenter/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.elevateab.com/elevate-helpcenter/mcp/mcp-overview.md).

# MCP Overview

Connect Elevate A/B Testing to an AI assistant using MCP, so you can ask about your experiment data in plain language instead of reading reports.

***

MCP lets you connect Elevate A/B Testing to an AI assistant so you can ask questions about your experiments in plain language. Instead of opening a report and reading it yourself, you ask your assistant something like "which experiments are running right now, and is any of them close to a winner," and it answers using live data from your store.

This page covers what the connection does, how to generate the API key every client needs, and how to pick the right setup path. If you already know which client you use, go straight to it:

| Your client    | Setup guide                                                                        | How it connects           |
| -------------- | ---------------------------------------------------------------------------------- | ------------------------- |
| Claude Desktop | [Claude Desktop](/elevate-helpcenter/mcp/connect-claude-desktop.md)                | Downloadable extension    |
| Claude Code    | [Claude Code and Cursor](/elevate-helpcenter/mcp/connect-claude-code-or-cursor.md) | Remote server, no install |
| Cursor         | [Claude Code and Cursor](/elevate-helpcenter/mcp/connect-claude-code-or-cursor.md) | Remote server, no install |

***

### What is MCP?

MCP stands for Model Context Protocol. It is an open standard that lets an AI assistant connect to an outside service and read data from it directly.

Without MCP, an AI assistant only knows what you paste into the chat. You would have to export a report, copy the numbers in, and ask your question. With MCP, the assistant queries Elevate A/B Testing itself, so it always sees your current data and you never copy anything.

The connection is read-only. Your assistant can look at your data and answer questions about it. It cannot create, start, pause, or end an experiment, and it cannot change any setting. It does not touch your Shopify store, your theme, or your storefront, and it has no effect on what visitors see.

### What you need before you start

| Requirement                    | Detail                                                                        |
| ------------------------------ | ----------------------------------------------------------------------------- |
| An Elevate A/B Testing account | MCP is available on every plan, with no upgrade required                      |
| A supported AI assistant       | Claude Desktop, Claude Code, or Cursor                                        |
| An API key                     | Generated in the app, covered below                                           |
| Desktop or laptop              | The extension and config file approaches both require a computer, not a phone |

Every plan reaches MCP the same way, through **Settings**, then **MCP**. There is no separate tier, no feature flag, and nothing to request.

You do not need a developer for the Claude Desktop path. It is a download and a paste. The Claude Code and Cursor path involves editing a config file, which is routine for anyone already using those tools.

### Where to find MCP in the app

Open **Settings**, then select **MCP** in the left sidebar. It sits below **Store Setup**.

The page has two sections: **API key** at the top, and **Connect your AI assistant** below it.

### Generate your API key

The API key is what proves to Elevate A/B Testing that a request is coming from your store. Every client uses the same key.

1. Go to **Settings**, then **MCP**.
2. Find the **API key** section at the top of the page.
3. Copy the key. It begins with `elv_`.

You get one key per store, and it works for every client listed on the page. If two people on your team use different assistants, they share the same key. If you manage several stores, each store has its own separate key.

Treat the key like a password. Anyone holding it can read your experiment data through the connection.

### Choose your client

The two paths are genuinely different, and the app will not accept the wrong one.

**Claude Desktop** installs a small extension on your computer. You download a file from the **Connect your AI assistant** section, double-click it, and paste your API key when Claude asks. The version number and supported operating systems are shown next to the **Download extension** button.

**Claude Code and Cursor** connect straight to the Elevate A/B Testing server over HTTPS. There is nothing to install. You add a short block to a config file, or run a single terminal command, using this server address:

```
https://elevateab.app/api/mcp
```

The two formats are not interchangeable. Claude Desktop does not accept the remote server configuration, so if you use Claude Desktop, use the extension.

### What the assistant can access

The Elevate A/B Testing MCP server gives your assistant 16 tools. Every one of them only reads data. None of them creates, changes, or deletes anything in Elevate A/B Testing or in Shopify.

| Tool                           | What it returns                                                                                                                                               |
| ------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `list_tests`                   | Your store's experiments, with optional filters                                                                                                               |
| `get_test`                     | Full details of one experiment, including every variation                                                                                                     |
| `get_test_results`             | Visitors, conversions, orders, revenue, profit, and lift for each variation, optionally narrowed by date range and the same filters as the results page       |
| `get_statistical_significance` | The significance status for each metric, recalculated on every request, with a verdict only once the experiment has enough data, the same as the results page |
| `get_test_timeseries`          | One metric over time for each variation, the data behind the performance chart                                                                                |
| `get_test_segmentation`        | Metrics by visitor segment: device, new or returning, organic or paid, and traffic source                                                                     |
| `get_test_products_breakdown`  | Results product by product, with revenue, orders, units, and conversion rate for each variation                                                               |
| `get_test_engagement`          | Session duration, where visitors go next, bounce rate, or click rates on elements                                                                             |
| `get_test_orders`              | Orders attributed to the experiment, with the same detail as the raw data export, including line items                                                        |
| `get_test_events`              | Individual events recorded for the experiment, with device, browser, page, location, and traffic source                                                       |
| `get_test_event_count`         | How many events match a given set of filters                                                                                                                  |
| `get_test_filter_options`      | The browsers, devices, countries, sources, pages, and UTM values that actually appear in the experiment's traffic                                             |
| `get_test_history`             | The experiment's activity timeline, such as started, paused, and settings changed                                                                             |
| `get_store_overview`           | Your store's plan, currency, monthly revenue, number of running experiments, and total experiments run                                                        |
| `get_store_conversion_funnel`  | Your store-wide funnel: page view, add to cart, checkout, purchase                                                                                            |
| `get_store_analytics`          | A store-wide report: web summary, period comparison, top pages, traffic sources, product performance, or cart analysis                                        |

You do not call these yourself. Ask your question in plain language and the assistant picks the tools it needs.

Generating, regenerating, and revoking your API key happen only in the Elevate A/B Testing app, never through the assistant.

### Is this safe?

Two things worth understanding before you connect.

**The connection is read-only.** Your assistant reads your Elevate A/B Testing data and nothing more. There is no tool that writes, so an assistant cannot end an experiment, change traffic allocation, or edit a setting even if you ask it to. It also has no access to your Shopify admin, your customer records, or your storefront code.

**Claude Desktop shows a security warning during install, and this is expected.** When you install the extension, Claude warns you that it can access your computer and that the developer is not verified by Anthropic. That warning appears for every locally installed extension, not just this one, because Claude cannot inspect what any extension does and so warns about what any program could in principle do.

The Elevate A/B Testing extension is three files, including one script of under 200 lines. It reads the messages Claude sends it, forwards them to Elevate A/B Testing over HTTPS, and returns the reply. You can read the source yourself before installing. The **Connect your AI assistant** section provides the exact command for macOS and Windows.

### Regenerate or revoke your key

Both actions are available in the **API key** section.

| Action         | What it does                                   | When to use it                                         |
| -------------- | ---------------------------------------------- | ------------------------------------------------------ |
| **Regenerate** | Issues a new key and invalidates the old one   | A team member leaves, or you think the key was exposed |
| **Revoke**     | Invalidates the key and leaves you without one | You want to switch the connection off entirely         |

Regenerating breaks every existing connection until each client is updated with the new key. If you use Claude Desktop, you do not need to reinstall the extension. Update the key in Claude under **Settings**, then **Extensions**, then **Elevate A/B Testing**.

### Terminology note

The app labels this feature and its data using the word "test" in places where this documentation uses "experiment." They refer to the same thing. When you ask your assistant about your experiments, it will find them.

### FAQs

#### Does connecting an AI assistant affect my running experiments?

No. The connection is read-only. It does not change traffic allocation, pause or start anything, or alter what visitors see on your store. Asking your assistant to make a change will not work, because no tool exists to do it.

#### Which plans include MCP?

All of them. MCP is available on every Elevate A/B Testing plan, and every plan reaches it the same way through **Settings**, then **MCP**.

#### Can my whole team use one key?

Yes. There is one key per store, and it is designed to be shared across your team and across clients.

#### I manage multiple stores. Do I need a key for each?

Yes. Keys are per store, so each store you manage has its own key and its own connection.

#### What happens if I regenerate my key?

Every client using the old key stops working until you paste in the new one. For Claude Desktop, update the key inside Claude rather than reinstalling the extension.

#### Can I use this on my phone?

No. Both setup paths require a desktop or laptop.

***

### Next steps

* [Claude Desktop](/elevate-helpcenter/mcp/connect-claude-desktop.md): Install the extension and connect
* [Claude Code and Cursor](/elevate-helpcenter/mcp/connect-claude-code-or-cursor.md): Connect to the remote server
* [Reports Overview](/elevate-helpcenter/analyze/reports/reports-overview.md): Understand the data your assistant is reading
* [Glossary](/elevate-helpcenter/getting-started/glossary.md): Plain English definitions of every metric your assistant may mention