Connect Claude Code or Cursor
Connect Claude Code or Cursor to the Elevate A/B Testing MCP server using a config file or a single terminal command, with no extension to install.
Claude Code and Cursor connect straight to the Elevate A/B Testing server over HTTPS. There is nothing to download and no extension to install. You add a short block to a config file, or run one terminal command, and the connection is live.
Once connected, you can ask about your experiments from inside your editor or terminal and get answers from your live store data.
If you use Claude Desktop, it connects a different way and will not accept the configuration on this page. See Claude Desktop.
Before you start
Claude Code or Cursor installed
Either client works with the same server
Your Elevate A/B Testing API key
Generated in Settings, then MCP
Access to Settings in Elevate A/B Testing
To copy the key
A plan
MCP is available on every plan, with no upgrade required
Step 1: Copy your API key
Open Elevate A/B Testing and go to Settings.
Select MCP in the left sidebar.
In the API key section, copy your key. It begins with
elv_.
You get one key per store, and the same key works for every client. If you or a teammate has already connected another assistant, reuse that key. Regenerating it disconnects everyone currently using it.
For more on how keys work, see MCP Overview.
Step 2: Add the server configuration
Both clients use the same server address and the same authorization header. Only the file location differs.
Claude Code
.mcp.json
Cursor
.cursor/mcp.json
Add this to the file, replacing YOUR_API_KEY with the key you copied:
If the file already contains other MCP servers, add the elevate entry inside the existing mcpServers object rather than creating a second one.
Step 2, alternative: Add it from the terminal
Claude Code can add the server without you editing a file. Run this, replacing YOUR_API_KEY with your key:
This produces the same result as editing the config file by hand. Use whichever you prefer.
Step 3: Confirm it is working
Restart your client, or start a new session. A session that was already running will not pick up the new server.
Ask about your experiments. Something simple works best for a first check, such as asking which experiments you currently have.
The answer should reflect your actual store data rather than general knowledge.
If nothing comes back, go to Troubleshooting below.
Keeping your key out of version control
The configuration above contains your API key in plain text. Anyone with the key can read your experiment data.
If the config file lives inside a repository, make sure it is not committed. Add it to your .gitignore before you add the key, not after, since a key that has already been committed stays in the history even once you remove it. If that happens, regenerate the key in Settings, then MCP.
If you regenerate your API key
Regenerating issues a new key and invalidates the old one immediately. Every client using the old key stops working until you update it.
For Claude Code and Cursor, update the key in the config file, then restart the client. If you added the server from the terminal, re-run the command with the new key.
Troubleshooting
The server does not appear in my client
Likely cause: The client was running when you edited the config, or the file is in the wrong place.
How to confirm: Check that the file is named exactly .mcp.json for Claude Code or .cursor/mcp.json for Cursor, and that the leading dot is present.
Fix: Correct the location if needed, then fully restart the client.
The connection appears but returns nothing
Likely cause: The API key is wrong, incomplete, or was regenerated or revoked.
How to confirm: Compare the key in your config file against the one shown in Elevate A/B Testing under Settings, then MCP.
Fix: Paste the current key, including the elv_ prefix, then restart the client. Check that YOUR_API_KEY was actually replaced and that the word Bearer and the space after it are still there.
My client reports a JSON error
Likely cause: The elevate block was pasted alongside an existing mcpServers object instead of inside it, leaving two objects at the same level.
How to confirm: Open the file and check there is only one mcpServers key.
Fix: Merge the entries so elevate sits inside the single mcpServers object, next to any other servers you have configured.
It works for me but not for a teammate
Likely cause: Config files are local. Your teammate needs their own copy.
Fix: Have them add the same configuration on their machine using the same API key. The key is shared across the store.
The connection stopped working with no changes on my end
Likely cause: Someone regenerated or revoked the shared key.
How to confirm: Compare your key against the one in Settings, then MCP.
Fix: Update your config with the current key and restart.
Claude Desktop rejected this configuration
Likely cause: Claude Desktop does not support the remote server format.
Fix: Use the extension instead. See Claude Desktop.
FAQs
Do Claude Code and Cursor need different keys?
No. One key per store, shared across every client and every person on your team.
Can I connect more than one store?
Add a second entry inside mcpServers with a different name and that store's key. [NEEDS VERIFICATION: whether the server supports concurrent connections from one client, and how the assistant is expected to distinguish between stores.]
Does this change anything on my store?
No. The connection is read-only. Your assistant can read your experiment data, but it cannot start, pause, or end an experiment, and it cannot change any setting. Your theme and storefront are untouched.
Which plans include this?
All of them. MCP is available on every Elevate A/B Testing plan.
Where exactly should .mcp.json live?
[NEEDS VERIFICATION: whether Elevate A/B Testing recommends project-level or user-level scope for Claude Code, since the app copy does not specify.]
What can I actually ask it?
[NEEDS VERIFICATION: blocked on the tool list. See MCP Overview.]
Next steps
MCP Overview: How keys, clients, and access work
Claude Desktop: Connect the desktop app instead
Reports Overview: Understand the data your assistant is reading
Last updated