Inspect network and page diagnostics

The network namespace controls request routing and capture. Diagnostics report console output, page errors, web vitals, accessibility results, and React tree data.

Focused snippets that start from browser are partial. Place them inside an active Browser context such as the request-routing example.

Route a request

from agentbrowser import Browser, RouteResponse

with Browser.launch() as browser:
    browser.network.route(
        "**/api/profile",
        response=RouteResponse(
            status=200,
            body='{"name":"Ada"}',
            content_type="application/json",
        ),
    )
    browser.open("https://example.com")
    body = browser.evaluate("fetch('/api/profile').then(response => response.text())")
    print(body)
    browser.network.unroute("**/api/profile")

Use either the typed response object or the shorthand status, body, content_type, and headers fields in one call. A supplied response takes precedence.

Set abort=True to stop matching requests. resource_type and resource_types narrow the route.

Inspect requests

requests = browser.network.requests(method="GET", status=200)
for request in requests:
    detail = browser.network.request_detail(request.id)
    print(request.url, detail.status)

requests(clear=True) returns matching summaries and clears the capture buffer. Filters include URL pattern, resource type, method, and status. network.credentials() sets HTTP authentication credentials.

Record an HTTP Archive

An HTTP Archive file records network request and response metadata.

with Browser.launch() as browser:
    browser.network.har_start(content="text")
    browser.open("https://example.com")
    har_path = browser.network.har_stop("trace.har")

print(har_path)

content="text" embeds text-like bodies up to 2 MiB each. "all" also embeds binary bodies as base64. "none" records metadata. One recording can embed up to 64 MiB.

HAR files can contain cookies, authorization headers, request data, and response bodies.

Read browser diagnostics

console = browser.diagnostics.console()
errors = browser.diagnostics.errors()
vitals = browser.diagnostics.vitals()

for message in console:
    print(message.level, message.text)

diagnostics.vitals() reloads the active page, waits three seconds for layout shifts and React effects, and returns a raw mapping. Time to First Byte (TTFB), First Contentful Paint (FCP), Interaction to Next Paint (INP), hydration, phase, and component timing values use milliseconds. Largest Contentful Paint (LCP) contains start time, rendered size, element name, and URL. Cumulative Layout Shift (CLS) contains a unitless score and layout-shift entries. The web.dev metric guides explain how to interpret these metrics.

React inspection requires the embedded React DevTools hook before application code runs. Set the engine feature before constructing the controller:

import os

from agentbrowser import Browser

os.environ["AGENT_BROWSER_ENABLE"] = "react-devtools"

with Browser.launch() as browser:
    browser.open("https://react.dev/")
    tree = browser.diagnostics.react_tree()
    print(tree)

diagnostics.react_tree(*, selector=None) returns React component data exposed through that hook. Calling it in a session launched without the feature raises BrowserError with the required setup.

Audit accessibility

Accessibility snapshots support agent interaction. An accessibility audit is a separate workflow that runs the embedded axe-core rules.

with Browser.launch() as browser:
    audit = browser.diagnostics.accessibility(
        "https://example.com",
        tags=("wcag2a", "wcag2aa"),
    )

for issue in audit.violations:
    print(issue.id, issue.impact, issue.node_count)

The tags select Web Content Accessibility Guidelines rule groups. Pass a URL to navigate before the audit or omit it for the active page. The selector must be valid CSS and match an element. Selector paths remain nested across frames and shadow roots. Audits require a CDP-capable browser.

Use Models and errors for the typed audit, network, and console results.