swiftbrowse

The AI browser for browser automation and end-to-end testing. A native engine (Rust, driving Chromium over the DevTools Protocol) with one contract across a test runner, this Python SDK, a TypeScript SDK, and an MCP server for AI agents. Standard library only.

📖 Documentation: https://steady-ember-82q7.here.now/

pip install swiftbrowse
npx playwright install chromium chromium-headless-shell   # a Chromium for SwiftBrowse to drive
swiftbrowse doctor                                       # check what this machine can run

Tests in code

import pytest
from swiftbrowse import launch

@pytest.fixture(scope="session")
def browser():
    with launch() as b:
        yield b

@pytest.fixture
def page(browser):
    with browser.new_session() as s:   # isolated: own cookies, storage, tabs
        yield s

def test_signs_in(page):
    page.navigate("http://localhost:3000/login")
    page.fill("Email", "ada@example.com")
    page.click({"by": "role", "role": "button", "name": "Sign in"},
               expect={"type": "url_contains", "value": "/dashboard"})
    page.expect({"type": "text_visible", "text": "Welcome, Ada"})

• Every engine operation is a method with its documented signature (navigate, click, fill, check, hover, select_option, upload_file, route, switch_tab, switch_frame, save_storage_state, …), with required arguments positional and the rest by keyword.
• A target is a string (accessible name, label, test id or text) or a locator (role, label, test_id, text, css, xpath). Several matches are an error, never a guess.
• Actions wait until the element is visible, enabled, not covered and not moving. expect assertions retry until they pass.
• A failed action or assertion raises SwiftBrowseError with the full result attached.
• Sessions attach to a warm background browser, so creating one per test is cheap.

Tests as JSON, with the built-in runner

{
  "name": "sign in",
  "tests": [
    { "name": "signs in", "steps": [
      { "op": "navigate", "url": "login" },
      { "op": "fill", "target": "Email", "value": "ada@example.com" },
      { "op": "click", "target": "Sign in", "expect": { "type": "url_contains", "value": "/dashboard" } }
    ] }
  ]
}

swiftbrowse test --base-url http://localhost:3000/ -j 4 --retries 2 --html report.html --trace retain-on-failure

The runner supports parallel workers, retries and flaky detection, sharding, saved logins, network mocking, tabs, frames, uploads, traces with a live DOM, video, and JUnit/HTML reports. On the project's benchmark suite it runs 2–3× faster than Playwright Test.

For AI agents

{ "mcpServers": { "swiftbrowse": { "command": "swiftbrowse", "args": ["mcp"] } } }

Learn more

• Getting started
• Python SDK
• Test files and the test runner
• Operations reference
• Issues and source

Platform wheels include the native binary (Linux x86_64/aarch64, macOS arm64/x86_64, Windows x64). Elsewhere, the pure-Python wheel uses a swiftbrowse on your PATH, or SWIFTBROWSE_BIN. Apache-2.0.