Headless browser

Shmastra ships Mastra's native AgentBrowser — a headless Chromium (Playwright) the agent can drive. Good for scraping, filling forms, and anything search snippets can't handle.

Example prompts

Give the research agent a browser so it can fetch and parse product pages.

Build a scraper agent that visits URLs I paste and extracts the main article text.

Let this agent navigate sites — I'll ask it to compare two checkout flows.

What happens

The widget adds browser: createAgentBrowser() to the agent:

import { createAgentBrowser } from '../shmastra'

export const myAgent = new Agent({
  // …
  browser: createAgentBrowser(),
})

Shmastra makes sure Playwright's Chromium Headless Shell is installed (the template ships npm run install-browsers); Docker images include it by default.

What the agent can do

• Navigate a URL, read the DOM.
• Click, type, wait for selectors.
• Take screenshots (which you can return as inline images via rich chat replies).
• Chain multiple pages together — login flows, searches, paginated scrapes.

Example conversations

• "Visit https://news.ycombinator.com/ and summarise the top five stories."
• "Compare prices for this product on Amazon and the manufacturer's site: …"
• "Fill in the signup form on example.com/signup with test data and tell me what errors it returns."

Tips

• Tell the agent to be polite. For public sites, ask it to respect robots.txt and throttle requests.
• Sites with CAPTCHAs won't work — the browser is real Chromium but no humans are behind it.
• Long-running scrapes belong in a workflow. Build a workflow whose step is an agent with a browser, and let it run in the background.