shalfa 4.0.0rc53

Python SDK for Stagehand

Stagehand SDK Client (Python)

Python SDK for controlling the Stagehand extension through ModCDP.

Run

From the repository root:

pnpm run build:artifacts
pnpm --filter @browserbasehq/stagehand-docs test:examples:python

For the CI smoke flow that exercises raw CDP, ModCDP, and GitHub summary output:

uv run python tests/integration/test_basic.py

Usage

import asyncio
import os

from stagehand import StagehandClient


async def main() -> None:
    client = StagehandClient(
        # Pass at most one launch source:
        # cdp_url="http://127.0.0.1:9222",
        # cdp_url="ws://127.0.0.1:9222/devtools/browser/<id>",
        # local_browser_launch_options={
        #     "executable_path": "/Applications/Chromium.app/Contents/MacOS/Chromium",
        # },
        # browserbase_session_create_params={
        #     "browserbase_api_key": os.environ["BROWSERBASE_API_KEY"],
        # },
    )

    await client.connect()

    client.on("Target.targetInfoChanged", print)

    print(await client.Browser.getVersion())
    print(await client.Target.getTargets())
    print(await client.Mod.evaluate(expression="chrome.runtime.id"))

    browser = client.browser
    page = await browser.new_page(url="https://example.com")
    page2 = await page.goto(url="https://browserbase.com")
    print(page2.url)
    print(await browser.pages(url=page2.url))

    body = await page.locate(css="body")
    print(await body.info())

    await client.close()


asyncio.run(main())

StagehandClient.connect() and StagehandClient.close() are async. Stagehand browser/page/locator methods are async. The compact client.* CDP command surface is synchronous because the Python ModCDP client owns a background websocket reader thread.

Launch Sources

connect() accepts exactly one effective launch source:

• cdp_url: an existing browser CDP endpoint. http://... URLs are resolved through /json/version; ws://... and wss://... URLs are used directly.
• local_browser_launch_options: launches local Chromium or Chrome Canary with CDP enabled. CHROME_PATH can provide the executable path. STAGEHAND_SDK_CDP_PORT fixes the debugging port for editor attach workflows.
• browserbase_session_create_params["browserbase_api_key"]: asks ModCDP to create a Browserbase session. The SDK uploads or reuses the packaged versioned extension artifact automatically; no extension ID configuration is required.

If no launch source is passed, connect() reads os.environ; it uses BROWSERBASE_API_KEY when present, otherwise it launches local Chromium/Canary.

Extension Paths

The default unpacked extension directory is the WXT output:

stagehand-extension/.output/chrome-mv3

The same directory contains the single reusable extension zip:

stagehand-extension/.output/chrome-mv3/stagehand-extension.zip

The prepared extension manifest uses the canonical stagehand-modcdp/background.js service worker path. The client trusts and waits for that service worker with globalThis.__stagehand_modcdp_ready === true.

Routing

Default client routes:

{
    "Mod.*": "service_worker",
    "Stagehand.*": "service_worker",
    "*.*": "service_worker",
}

Default extension service-worker routes:

{
    "Mod.*": "service_worker",
    "Stagehand.*": "service_worker",
    "*.*": "loopback_cdp",
}

Stagehand SDK routes are fixed: Stagehand and ModCDP commands go through the extension service worker, and native CDP commands go from the service worker to loopback CDP.

Debugging

The checked-in VS Code tasks set:

STAGEHAND_DEBUG_WAIT_FOR_ENTER=1
STAGEHAND_SDK_CDP_PORT=9231

The smoke test at tests/integration/test_basic.py prints [debug] Browser ready. Press Enter to continue. after the extension is connected, so DevTools can inspect chrome-extension://*/stagehand-modcdp/background.js before the flow continues.