Browser

An agent can start your app. Without a browser it cannot see it. It has to ask you whether the page renders, whether the console is clean and whether the request went out. When you are not there, the run stops and waits for you.

The browser feature puts a headless Chromium into the workbench, together with the Playwright MCP server that drives it. The agent opens your app, reads the page, clicks and types, and checks the console and the network requests on its own. The same browser also runs your project’s own Playwright tests.

Add it

monoceros add-feature acme browser

Then apply the workbench:

monoceros apply acme

You don’t need to add an MCP server. apply registers one called playwright with every AI agent in the workbench. It works like an entry in the mcpServers: block. A workbench without an agent gets only the browser, for your tests.

How the agent uses it

The browser runs inside the container, next to your dev server. So the agent opens your app at http://localhost:<port>, with the port your dev server listens on. It does not use the .localhost address you open on your own machine.

Ask for the check in plain words:

Start the dev server, open the sign-up page and check that the form submits without console errors.

The workbench briefing tells the agent what to do. The agent reads the page and acts on it. Before it calls the page working, it checks the console messages and the network requests.

If you ask the agent for a screenshot, you find it in /tmp/playwright-mcp in the container.

Your project’s Playwright tests

Ask the agent for a test the same way you ask it for code:

Write a Playwright test for the sign-up flow and run it.

The agent sets up what the project needs, writes the test and runs it. Every action it takes in the browser shows the Playwright code behind it. When the agent writes a test for a flow it just clicked through, it starts from that code.

Running the tests yourself

They are normal Playwright tests. You can also run them yourself, from the container shell:

npx playwright test

The tests need a browser build that matches your Playwright version. If the agent set up the tests, it installed that build already. If not, install it once in your project:

npx playwright install --only-shell chromium

Leave out --with-deps. The workbench already has the system libraries. The workbench also keeps the download when you apply. You only install it again when you update Playwright in your project.

What it costs

Chromium and its libraries add a few hundred megabytes to the workbench. apply installs them when it builds the workbench, so the first apply takes longer.

The browser is headless, so there is no window. To look at the app yourself, open it in your own browser.