Tests written by an AI agent

Connect your AI assistant to your account and it writes the tests for you: it studies the site, designs the scenarios and creates them in your account - you just review and run them.

Start free

Ask in plain words

"Write tests for the checkout flow" - the assistant works out which steps are needed and in what order. You do not need to know the scenario format.

Selectors are not invented

The assistant opens your page and picks selectors with the same engine the visual recorder uses - with fallbacks in case the markup changes.

Your own tests stay untouched

The assistant can see every scenario so it will not create duplicates, but it can only modify the ones it created itself. Anything you recorded stays as it is.

How to set it up

  1. Create a token in your account: Profile → API tokens, purpose "AI agent (MCP server)".
  2. Add one section with the address and the token to your AI client configuration - nothing to install, the server runs on our side.
    .mcp.json
    {
  "mcpServers": {
    "testmw": {
      "type": "http",
      "url": "https://testmw.com/mcp",
      "headers": { "Authorization": "YOUR_TOKEN" }
    }
  }
}

    Your client will show whether the server connected: in Claude Code, for example, run /mcp - testmw should be in the list.

  3. Ask the assistant to write tests for your site, then review the result in your account.

Scenario steps: Every scenario step - the agent picks the right ones itself

What the assistant can do

  • See your plan - how many tests you can still create, the step limit, which run intervals are available.
  • Know every TestMW step - clicks, typing, assertions, waits - with hints on what each step expects as its value.
  • Study your site - open your pages and pick durable element selectors.
  • Create and update tests - the finished scenario shows up in your account right away.
  • Run tests and read results - up to fixing a failed step on its own.

How to use it

Just ask the assistant in plain words:

  • Check my TestMW plan and tell me how many more tests I can create
  • Write a checkout test for shop.example.com
  • Run test 42 and show me what failed
  • Fix the selector in the test that started failing yesterday

The assistant decides which tools to call - you do not need to name them.

Where the selectors come from

Whether a test survives a markup change depends on the selectors its steps are recorded with. The assistant does not invent them:

  • If the assistant has a browser - it opens your page and picks selectors with the same engine as the TestMW visual recorder - with fallbacks in case the markup changes.
  • If it has no browser - steps are recorded by meaning (the "Checkout" button), and the actual element is resolved when the test runs. Such tests work too, but they are less durable.

The page is opened by your assistant's browser - with its own cookies and its own IP. If you explored the site logged in and the test later runs from a clean slate, the scenario may fail: ask the assistant to run the recorded test and confirm it passes.

If something does not work

Problem What to do
The server does not show up in the list check that the config says "type": "http" and the address ends with /mcp.
Access error the token is wrong or revoked, or your plan does not include the AI agent (including an expired plan). Make sure the token was copied whole, with no stray spaces.
The assistant cannot update a test it only changes tests it created itself. Scenarios you recorded by hand or with the recorder are protected from overwriting - ask it to create a new test.

Still stuck - contact support from your account.

All features

Try it on your own site

Give us a link - we handle the rest

Start free Pricing