For agents

A/B testing your agent can set up itself

Your coding agent creates the Mida account, installs the tag or the SDK, and gets a first test ready to launch. Your part is one 6-digit code from your inbox.

Sign up yourself
  • No signup form. One emailed code, read to your agent.
  • Free up to 100,000 tracked users a month. No credit card.
  • Every test starts as a draft. You decide what goes live.
agent ~ mida setup 
Set up Mida A/B testing in this project.

1. Read https://mcp.mida.so/auth.md and https://docs.mida.so/llms.txt first.
2. Ask me which email to register and whether my visitors are mostly in the EU. Show me https://www.mida.so/terms before you register.
3. Run: npx mida-cli register --email MY_EMAIL --accept-terms
   Add --region eu for the EU. Mida emails me a 6-digit code. Ask me for it, then run: npx mida-cli claim --code CODE
4. Run npx mida-cli snippet and add the tag to the head of my site layout, exactly as printed. If this is a backend or an app, install the Mida SDK for this stack instead.
5. After I deploy, run npx mida-cli doctor --wait to confirm data is arriving.
6. Suggest one test worth running here. Create the goal and the experiment as a draft, then run npx mida-cli check TEST_ID and fix what it reports.
7. Do not set anything live. Show me the preview link and ask first.

Paste it into Claude Code, Cursor, Codex, Windsurf or any agent that can run a command in your project.

How it works

From empty repo to a test ready to launch

Three steps, all from the terminal your agent already works in. You can run them yourself too. It is the same commands either way.

01

Sign up

The agent registers your email. Mida sends you a 6-digit code, you read it out, and the keys land in .env. The API key is never printed.

$ npx mida-cli init --email you@acme.com --accept-terms
Code from the email: 482913
Account created. Project key: 1Avp2Dgn... (us)

02

Install and check

It adds the tag to your layout, or the SDK to your backend, then waits for the first real visitor so you know the install worked before anything is built on it.

$ npx mida-cli snippet
$ npx mida-cli doctor --wait
Install health: no_test_collecting
Visitors in the last 7 days: 2

03

Build a draft, then ask you

It creates a goal and a draft test, runs the same setup checks the dashboard runs, and hands you a preview link. Launching is your call.

$ npx mida-cli check 1234
Experiment 1234: ready. 13 checks passed.

Interfaces

Work in the tool your agent likes best

One account, four ways in. Mix them: sign up from the CLI, build over the API, read results in chat.

CLI

Sign up, check the install, and call any API endpoint as a command, without leaving the shell.

npx mida-cli init

REST API

Experiments, goals, results and settings over plain HTTP, with an index written for agents.

docs.mida.so/llms.txt

MCP server

Connect Claude, Cursor or ChatGPT to your account and run the whole loop in conversation.

mcp.mida.so/mcp

SDKs

Server-side tests in your own code: Node, Python, Go, PHP, Ruby and more. Now on the free plan.

npm install mida-node

Free to start

Build for free. Pay when you need more.

The account your agent creates is a normal Mida account on the free plan. No credit card, and nothing to migrate if you upgrade later.

100,000

Tracked users a month

Enough to run real tests on a real site before you spend anything.

3 live tests

At a time, with unlimited goals

Client-side and server-side. Drafts are not counted, so your agent can prepare as many as it likes.

Everything here

CLI, API, MCP server and SDKs

The agent tooling is not a paid add-on. It is how the product works.

See what the paid plan adds

FAQ

Questions before you hand it over

What does my agent get access to?

One API key for one project. The CLI writes it to .env, adds that file to .gitignore, and never prints it. You can revoke the key in the dashboard under Settings, API, like any other key.

Does it ever see my password?

No. The account is created without one. The only thing you give the agent is the 6-digit code from your inbox, which works once and expires after 15 minutes. When you want to open the dashboard, you set a password yourself from the "Forgot password" page.

Can it launch a test without asking me?

Every test is created as a draft. Through the MCP server, launching needs your explicit yes, and the server refuses the call without it. Through the REST API a key can change a test's status, so the setup prompt and the Mida skill both tell the agent to stop and ask first. If you want the stricter of the two, connect over MCP.

I already have a Mida account. What happens?

Nothing is sent, and nothing is created. For privacy, Mida gives the same answer whether or not an address has an account. If no code arrives within a few minutes, sign in to the dashboard and create an API key under Settings, API, or connect the MCP server, which signs you in through your browser.

Which region should I pick?

If your visitors are mostly in the EU, add --region eu so the account and its data live in the EU. Otherwise leave it out for the US. An account cannot change region later, so answer this one yourself.

Which agents does this work with?

Any agent that can run a shell command or make an HTTP request in your project: Claude Code, Cursor, Codex, Windsurf, Gemini CLI and others. If your agent cannot run commands, the same two steps are plain HTTP calls, documented in the API reference.

If you are the agent

Start with these. They are written for you and kept current.

Hand your agent the first test

Copy the prompt, paste it into your agent, and read it one code. You get a draft test to review, not a setup guide to read.

See what agents do with Mida

Free plan. No credit card.