OpenAI integration with Playwright

Use your existing Playwright code with a remote Browser Use Cloud browser. OpenAI runs the model loop. A small adapter exposes Playwright actions as tools. This example exposes read_page, click, type, and select_option. OpenAI shared the small adapter used here; it is not a general Playwright driver. Playwright methods become model tools only when the adapter implements them. For all 28 browser tools and Bash, use the Browser Use library driver.

Connect your remote browser

Replace local browser startup with these lines:

session = await cloud.browsers.create(timeout=5)
browser = await playwright.chromium.connect_over_cdp(session.cdp_url)
page = await browser.contexts[0].new_page()

Chrome DevTools Protocol (CDP) connects Playwright to your remote Chromium browser. Your application runs Playwright and the OpenAI SDK. Browser Use Cloud hosts the browser.

Run the example

You need uv and API keys for Browser Use and OpenAI. You need no local Chromium installation or open-source Browser Use library.

mkdir openai-playwright
cd openai-playwright
uv init --bare --python 3.12
uv add openai==3.31.0 browser-use-sdk==3.11.3 playwright==1.63.0

export BROWSER_USE_API_KEY="your-browser-use-key"
export OPENAI_API_KEY="your-openai-key"
export OPENAI_MODEL="your-model"

curl -fsSL https://raw.githubusercontent.com/browser-use/sdk/e4048eafa0dc61d8afc112a2d2d09f97022e1677/examples/openai/playwright_browser.py -o playwright_browser.py
curl -fsSL https://raw.githubusercontent.com/browser-use/sdk/e4048eafa0dc61d8afc112a2d2d09f97022e1677/examples/openai/playwright_cloud.py -o playwright_cloud.py
uv run python playwright_cloud.py

Get your Browser Use key and OpenAI key. Choose an OpenAI model that supports Responses function tools. The model reads example.com, clicks its link, and summarizes the destination. The script stops the Cloud browser when the task ends.

Complete example

playwright_cloud.py

"""Run OpenAI browser actions through Playwright and Browser Use Cloud."""

import asyncio
import os

from browser_use_sdk.v4 import AsyncBrowserUse
from openai import AsyncOpenAI
from openai.lib.browser import AsyncBrowserToolset
from openai.lib.responses_tool_runner import AsyncResponsesToolRunner
from playwright.async_api import async_playwright

from playwright_browser import PlaywrightBrowser


async def confirm(context):
    return True


async def main() -> None:
    async with (
        AsyncBrowserUse() as cloud,
        AsyncOpenAI(timeout=180, max_retries=5) as client,
        async_playwright() as playwright,
    ):
        session = await cloud.browsers.create(timeout=5)
        browser = None
        try:
            if not session.cdp_url:
                raise RuntimeError("Browser Use did not return a CDP URL.")
            browser = await playwright.chromium.connect_over_cdp(session.cdp_url)
            page = await browser.contexts[0].new_page()
            await page.goto("https://example.com", wait_until="domcontentloaded")
            driver = PlaywrightBrowser(page)
            tools = AsyncBrowserToolset(
                driver, members=["read_page", "click", "type", "select_option"], confirm=confirm
            )
            runner = AsyncResponsesToolRunner(browser=tools)
            try:
                response = await runner.run(
                    client,
                    {
                        "model": os.environ["OPENAI_MODEL"],
                        "input": "Read the page, click Learn more, and summarize the destination.",
                        "store": False,
                    },
                    max_steps=10,
                )
                print(response.output_text)
            finally:
                runner.close()
                tools.close()
                await driver.aclose()
        finally:
            try:
                if browser is not None:
                    await browser.close()
            finally:
                await cloud.browsers.stop(session.id)


if __name__ == "__main__":
    asyncio.run(main())

The adapter keeps element references from each page read. Read again after navigation or when an element reference becomes stale. This small example supports one page and a limited set of visible form controls. The confirmation callback approves actions for this example. Use tasks you authorize, and change the callback for your application’s permissions.

Choose your integration

Your application	Integration
Needs all browser tools, JavaScript, coordinate actions, file tools, and Bash	Browser Use driver
Already uses Playwright and needs a small adapter	This example

Both integrations use Browser Use Cloud infrastructure. The adapter determines which actions the model can call.

Get Started

Pillars

More

OpenAI integration with Playwright

Connect your remote browser

Run the example

Complete example

Choose your integration

​Connect your remote browser

​Run the example

​Complete example

​Choose your integration

Connect your remote browser

Run the example

Complete example

Choose your integration