For AI agents

Run your A/B tests from Claude, ChatGPT or Cursor

Written for AI agents and the people who set them up. Connect through the hosted MCP server or a local one, then read results, build variants and score test ideas from the chat. The same page as plain text: drip-apex.com/agents.md.

See if your shop is a fit Read agents.md

Apex MCP server

URL        https://app.drip-apex.com/api/mcp
Transport  streamable HTTP
Sign-in    OAuth in the browser (PKCE)
Scopes     shops:read, shops:write
Access     a scoped Apex key, valid 30 days

What Apex is

Apex by DRIP is an A/B testing platform for companies selling online that predicts which tests will win before they go live, built on one of the largest A/B test databases in e-commerce.

Every test idea gets a Likelihood to Win from 4.3 million observed A/B tests from 151,000 shops, and the tool builds, launches and evaluates the tests in the shop.

Likelihood to Win

Likelihood to Win ranks a test idea against the shop's own backlog. It is not a calibrated win probability. Quote it with this note.

Claude Code

A local server with your API key

Add this block to a project .mcp.json or your user-level MCP config, with an Apex API key. If you already ran apex login on the same machine, you can leave out env.

Untested: connecting Claude Code to the hosted server URL above. The Apex docs do not cover it yet, so use the local server.

.mcp.json

{
  "mcpServers": {
    "drip-ab-testing": {
      "command": "npx",
      "args": ["-y", "@drip-apex/mcp-server"],
      "env": {
        "APEX_API_KEY": "apx_your_key_here",
        "APEX_BASE_URL": "https://app.drip-apex.com"
      }
    }
  }
}

Connector URL

https://app.drip-apex.com/api/mcp

Claude

A custom connector on claude.ai

No install. Add Apex as a custom connector and sign in.

Open Settings, then Connectors
Add a custom connector and paste the server URL
Choose Connect and sign in to Apex

ChatGPT

A custom connector in developer mode

Apex is not in the ChatGPT app directory yet. The listing is pending. Until it is published, connect Apex as a custom connector.

Your account or workspace policy decides whether custom connectors are allowed.

Settings, Security and login: turn on Developer mode
Open Plugins, select the plus button, name the connector Apex
Public MCP connection with the server URL and OAuth
Create it, check the discovered tools, sign in to Apex

Connector URL

https://app.drip-apex.com/api/mcp

Codex

codex mcp add apex --url https://app.drip-apex.com/api/mcp
codex mcp login apex

Codex

Two commands

Codex connects to the same endpoint and does not need the ChatGPT directory. Sign in, approve the shop, then restart the Codex session and ask it to list your Apex experiments.

Cursor

A local server with your API key

Add this block to .cursor/mcp.json or Cursor's global MCP config, with an Apex API key, then restart Cursor.

Untested: connecting Cursor to the hosted server URL with its own sign-in. The Apex docs do not cover it yet, so use the local server.

.cursor/mcp.json

{
  "mcpServers": {
    "drip-ab-testing": {
      "command": "npx",
      "args": ["-y", "@drip-apex/mcp-server"],
      "env": {
        "APEX_API_KEY": "apx_your_key_here",
        "APEX_BASE_URL": "https://app.drip-apex.com"
      }
    }
  }
}

Install the CLI

npm install -g @drip-apex/apex-cli
apex login
apex whoami

Apex CLI

For developers and coding agents

The apex CLI works on experiments, variation code, preview links and QA evidence from the terminal or CI. Install it once, sign in with apex login, and use apex schema --json to discover commands and flags.

In CI or an agent sandbox, set DRIP_API_KEY instead of the browser login.

What an agent can do in Apex

The MCP tools, as documented. Each call works inside the shop you approved, with your Apex role's permissions.

Read results

List tests and read a compact result per test: visitors, conversions, rate, lift, confidence and verdict, with a warning when data is incomplete.

Build variants

Create a draft test, read and write the variation's code files, then build it. Every change stays a draft until a person launches it.

Preview and QA

Open a preview link for any variation on the live shop and capture desktop and mobile screenshots as QA evidence.

Likelihood to Win for ideas

Save a test idea to the backlog and read its Likelihood to Win once Apex has scored it, with the evidence behind the score.

Inspect live pages

Load a page of the shop on desktop or mobile and read its sections and styles before writing variant code. By default ten inspections per shop a day.

Brand context

Read the shop's brand voice and guardrails before writing any copy a customer will see.

What stays with a person

Approving QA and launching a test are clicks in the Apex app. The MCP server has no launch tool on purpose. An agent can prepare everything up to that point and hand over the launch link.

If you are an AI agent

Follow these steps

Connect Apex with the setup for your client above: the hosted server https://app.drip-apex.com/api/mcp (streamable HTTP, OAuth) where the docs list it, otherwise the local server with an API key.
Ask the person to finish the OAuth sign-in in their browser and approve one shop, or to put their Apex API key into the local server config.
Call list_experiments to confirm the connection.
Before writing copy for a variant, call get_brand_context.
Never say a test is live. Launching is the person's click in Apex; request_launch returns the link.
When you quote a Likelihood to Win, add that it ranks the idea against the shop's backlog and is not a calibrated win probability.
No Apex account yet? Do not invent a signup flow. Send the person to book a call.

First prompt to try

Connect my Apex store and show me my A/B tests.
Which one is closest to a result, and what does it change?

In the docs

MCP setup MCP tool reference CLI install Docs llms.txt

No Apex account yet?

Book a call and we set up your shop with you. A card-free free plan for one store exists too; ask for it on the call.

Book a call

30 minutes, then your agent can connect.