Short answer: it’s not a good fit. Playwright needs system-level browser binaries and OS libraries that Prefect’s Managed work pool does not allow you to install or customize. Managed pools let you add pip packages, but you cannot bring a custom Docker image or apt/yum packages, nor run pre/post shell steps. In practice, Playwright will generally fail there even if

pip install playwright

succeeds. Recommended approaches: - Docker work pool with a Playwright base image (best): Build an image from

<http://mcr.microsoft.com/playwright/python|mcr.microsoft.com/playwright/python>

, add your code/deps, and run in a Docker work pool. - Kubernetes work pool: Same idea—use a Playwright-enabled image. - Process/EC2 worker: Run a pull worker on infrastructure where Playwright is already installed. Example (Prefect 3.x) using a Docker work pool: - Dockerfile

FROM <http://mcr.microsoft.com/playwright/python:latest|mcr.microsoft.com/playwright/python:latest>
# Optional: pin Python version tag above for stability
RUN pip install prefect playwright
# If you need browsers explicitly (the base image already has them)
# RUN python -m playwright install --with-deps chromium
# Add your code
WORKDIR /app
COPY . /app

- Deploy your flow to the Docker work pool (Python API)

from prefect import flow

@flow
def my_playwright_flow():
    # your Playwright code here
    pass

# Assuming you've created a Docker work pool named "docker-pool"
my_playwright_flow.from_source(".").deploy(
    name="playwright-deploy",
    work_pool_name="docker-pool",
    job_variables={"image": "your-registry/your-playwright-image:tag"}
)

If you absolutely must try Managed, the only lever you have is

pip_packages

and hoping Playwright can download a browser at runtime—however, the required system libraries won’t be present, so this is not a supported or reliable path. Docs and references: - Managed work pools overview and limits: Managed work pools (Prefect 3) - Work pools (Docker/Kubernetes) management: Manage work pools - Background on work pools and when to use pull vs. managed: Work pools and workers blog If you share which environment you prefer (Docker, K8s, or a process worker) and where you host images, I can give you a tailored Dockerfile and deployment steps.

Short answer: there isn’t a fully equivalent “pure Python” browser automation library that will run natively in a Managed work pool without system deps. Anything that actually renders a browser (Playwright, Selenium, Puppeteer/pyppeteer, undetected-chromedriver, etc.) needs OS-level binaries and libs — which Managed pools don’t allow. What does work in a Managed pool - Use a remote/headless browser service (SaaS or self-hosted) and control it from Python. Your flow only needs pip-installable clients; the browser runs elsewhere. - Playwright “remote” via WebSocket/CDP to a hosted service (e.g., a Browserless-compatible endpoint) or your own Chrome/Chromium with remote debugging. - Selenium RemoteWebDriver to a cloud grid (e.g., BrowserStack, Sauce Labs) or your own Selenium Grid. Examples you can run in a Managed pool (pip-only): - Playwright connecting to a remote browser

from prefect import flow
from playwright.sync_api import sync_playwright

@flow
def run_playwright():
    with sync_playwright() as p:
        # Browserless or your own Chrome with remote debugging enabled
        browser = p.chromium.connect_over_cdp("<wss://your-browser-host.example/ws?token=...>")

        context = browser.contexts[0] if browser.contexts else browser.new_context()
        page = context.new_page()
        page.goto("<https://example.com>")
        print(page.title())

Deployment job variables for Managed pool could include:

{
  "pip_packages": [
    "playwright"
  ]
}

- Selenium to a cloud/grid

from prefect import flow
from selenium import webdriver
from <http://selenium.webdriver.common.by|selenium.webdriver.common.by> import By

@flow
def run_selenium():
    options = webdriver.ChromeOptions()
    driver = webdriver.Remote(
        command_executor="https://<user>:<key>@hub.browserstack.com/wd/hub",
        options=options,
    )
    driver.get("<https://example.com>")
    print(driver.title)
    driver.quit()

Deployment job variables:

{
  "pip_packages": [
    "selenium"
  ]
}

Notes - Store tokens/keys for remote services as Prefect variables or environment variables, not in code. - Managed pools support pip packages but not system packages or custom images. If you ever need to run browsers locally, switch to a Docker/Kubernetes work pool. Docs on Managed pool capabilities/limits: Managed work pools (Prefect 3) If you tell me whether you prefer a hosted service or self-hosting a small browser container, I can share a tailored setup and deployment snippet.