Puppeteer

Drive a headless Chromium browser — navigate, screenshot, click, type, and run JS in the page.

Requires Growth plan or above.

Enable Puppeteer

Stateless — no credentials required. Toggle Puppeteer on at /dashboard/connections. Each call shares a per-app browser session within the MCP runner.

Example

Navigate to a URL and take a full-page screenshot.

app/api/page-shot/route.ts

export async function POST() {
  const headers = {
    'x-api-key': process.env.LEASH_API_KEY!,
    'content-type': 'application/json',
  }

  await fetch('https://leash.build/api/integrations/puppeteer/puppeteer_navigate', {
    method: 'POST',
    headers,
    body: JSON.stringify({ url: 'https://leash.build' }),
  })

  const res = await fetch('https://leash.build/api/integrations/puppeteer/puppeteer_screenshot', {
    method: 'POST',
    headers,
    body: JSON.stringify({
      name: 'leash-homepage',
      fullPage: true,
    }),
  })
  const { data } = await res.json()

  // data =>
  // {
  //   name: 'leash-homepage',
  //   data: 'iVBORw0KGgoAAAANSUhEUgAAB...', // base64 PNG
  //   width: 1280,
  //   height: 3214
  // }

  return Response.json({ screenshotBytes: data.data.length })
}

Available tools

puppeteer_navigate — load a URL.
puppeteer_screenshot — capture the current page (base64 PNG).
puppeteer_click — click an element by CSS selector.
puppeteer_fill — type into an input or textarea.
puppeteer_select — pick a <select> option.
puppeteer_hover — hover an element.
puppeteer_evaluate — execute JS in the page context and return the result.

Common gotchas

Selectors must already be in the DOM. For SPAs, navigate then puppeteer_evaluate a waitForSelector before clicking.
No persistent cookies. Each app session starts fresh; perform login in-flow rather than relying on stored state.
Screenshots can be large. Full-page captures on long pages routinely exceed 1MB; gate by URL/page-type before storing.

See the Puppeteer MCP server source for the underlying provider docs.

Other languages: see SDK overview.