SDKs

Python SDK

pip install mrbrowser gives you the full engine with a Pythonic API — designed to feel native inside pytest.

# Basic usage

quickstart.py

python

from mrbrowser import MrBrowser

with MrBrowser() as browser:
    page = browser.open("https://shop.example.com")
    page.type("the search field", "mechanical keyboard")
    page.click("the search button")
    page.click("the first product card")
    assert "Add to cart" in page.get_html()

# Pytest integration

Share one Memory Engine database across your suite so fingerprints accumulate run over run — that history is what powers self-healing in CI.

test_checkout.py

python

# conftest.py
import pytest
from mrbrowser import MrBrowser

@pytest.fixture
def browser():
    with MrBrowser(host="localhost", port=7331) as b:
        yield b

# test_checkout.py
def test_checkout_survives_redesign(browser):
    page = browser.open("https://shop.example.com")
    page.click("Add to cart")
    page.click("the checkout button")
    assert page.assert_text("Order summary")
    price = page.extract_text("the order total")
    assert price.startswith("$")

Tip

Commit .mrbrowser/memory.db to your repo (it's a compact SQLite file). Fresh CI runners then start with the full fingerprint history instead of a cold cache.

# Semantic extraction

extract.py

python

page = browser.open("https://billing.corp.com/invoices")

# Extract a specific field by plain-English description
balance = page.extract_text("the account balance")

# Inspect all elements on the page (useful for debugging)
elements = page.inspect(visible_only=True)
for el in elements:
    print(el["text"], el["role"], el["selector"])

# Cookie management
cookies = page.get_cookies()
page.set_cookies([{"name": "session", "value": "abc123"}])

# Run JS
result = page.execute_js("return document.title")

Warning

extract() resolves intents per field, per row. On tables with 1000+ rows, pass batch=True to fingerprint the row structure once — otherwise extraction time grows linearly.