How to scrape RSS and Atom feeds with Playwright

To scrape an RSS or Atom feed with Playwright, do three things in order: read the feed URL from the page’s <link rel="alternate"> tag instead of guessing a path, navigate to that URL in the same browser and read page.content() so the XML comes back past the same protection as the page, then parse the result as XML and branch on the root element (<rss> versus <feed>), because the two schemas share almost nothing.

A feed looks like the easy case. It is structured XML, it is meant to be read by machines, and for years the advice was to skip the browser entirely and fetch the URL with a plain HTTP client. That advice quietly stopped working, because the feed endpoint now usually sits behind the same edge protection as the HTML pages it belongs to, and a bare client fetching the XML gets a challenge instead of a document.

This page covers the three parts people get wrong: finding the feed URL in the first place, fetching the XML in a way that survives the protection, and parsing a format that is really two incompatible formats wearing the same name.

Find the feed URL from the page, do not guess it

Read the feed URL from the page itself instead of guessing a path: a site that publishes a feed advertises it in the document head with a <link rel="alternate"> element whose type is application/rss+xml or application/atom+xml.

There is no fixed path for a feed. /rss, /feed, /atom.xml, /index.xml and a dozen others are all in use, and guessing wastes requests against exactly the kind of endpoint that counts requests. Read that attribute instead:

from urllib.parse import urljoin
from invisible_playwright import InvisiblePlaywright

FEED_SELECTOR = (
    "link[rel='alternate'][type='application/rss+xml'], "
    "link[rel='alternate'][type='application/atom+xml']"
)

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com/blog")

    href = page.get_attribute(FEED_SELECTOR, "href")
    if href is None:
        raise SystemExit("no feed advertised on this page")

    # the href is often relative, so resolve it against the page URL
    feed_url = urljoin(page.url, href)
    print("feed:", feed_url)

page.get_attribute returns the first match, which is what you want when a site advertises both an RSS and an Atom version of the same content. If you need every feed on the page, page.query_selector_all(FEED_SELECTOR) gives them all and you can read href from each.

Fetch the XML through the browser, not a side client

Fetch the feed by navigating to its URL in the same browser that loaded the page and reading page.content(), not by handing the URL to a separate HTTP client. This is the part that trips up feed scrapers built the old way. Once you have the feed URL, the instinct is to hand it to a separate HTTP library, because it is “just XML”. On a protected site that separate request is a fresh visitor with none of the context the browser accumulated, and it draws the challenge the page navigation did not.

Navigate to the feed URL in the same browser instead, and read the served document with page.content():

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com/blog")

    href = page.get_attribute(FEED_SELECTOR, "href")
    feed_url = urljoin(page.url, href)

    page.goto(feed_url)
    xml_text = page.content()   # the served feed, past the same edge protection

The browser is a real browser: a Firefox patched at the C++ level, driven by stock Playwright, presenting a consistent TLS handshake and fingerprint that a stripped-down client cannot. The feed comes back because the request that asked for it is indistinguishable from the request that asked for the page. If you want to keep the raw response bytes exactly as served rather than the serialized DOM, capturing the response body directly with page.on("response") is the companion technique, and it avoids any question of what the browser did to the document on the way in.

One caveat worth stating plainly, because it is the honest one: whether the feed loads at all is downstream of whether the session looks real. If the HTML site is challenging you, the feed will challenge you too, and the fix is the same fix as for any blocked page rather than anything feed-specific. That whole order of operations is the checklist for scraping without getting blocked, and it applies here unchanged.

Parse it as XML, and branch on the schema

Parse the feed with a standard XML parser and branch on the root element: <rss> is an RSS document, <feed> is Atom, and the two share almost nothing structurally. “RSS feed” is a colloquial name for both, and code that assumes one silently returns nothing on the other. The differences that matter:

  • RSS wraps items in a <channel>, each item is <item>, the date is <pubDate> in RFC 822 format, and the body is <description>.
  • Atom has no channel, each item is <entry>, the date is <updated> in ISO 8601 format per RFC 4287, the body is <content> or <summary>, and the link is an attribute (<link href="...">) rather than element text.

So the first thing to do after parsing is look at the root element and decide which world you are in. Do not sniff the URL or the content type, read the tag. This is the same root-element branch used when scraping a sitemap.xml through the browser, where <sitemapindex> and <urlset> split apart the same way:

import xml.etree.ElementTree as ET

ATOM = "{http://www.w3.org/2005/Atom}"


def parse_feed(xml_text):
    root = ET.fromstring(xml_text)
    tag = root.tag.split("}")[-1]   # strip any namespace prefix

    if tag == "rss":
        channel = root.find("channel")
        return [_rss_item(node) for node in channel.findall("item")]

    if tag == "feed":   # Atom's root element
        return [_atom_entry(node) for node in root.findall(ATOM + "entry")]

    raise ValueError("unrecognized feed root element: " + tag)


def _rss_item(item):
    enclosure = item.find("enclosure")
    return {
        "title": item.findtext("title"),
        "link": item.findtext("link"),
        "date": item.findtext("pubDate"),
        "body": item.findtext("description"),   # CDATA HTML, already unwrapped
        "media": enclosure.get("url") if enclosure is not None else None,
    }


def _atom_entry(entry):
    # Atom's link is an attribute; prefer rel="alternate" for the article link
    link = None
    media = None
    for node in entry.findall(ATOM + "link"):
        rel = node.get("rel", "alternate")
        if rel == "alternate":
            link = node.get("href")
        elif rel == "enclosure":
            media = node.get("href")

    body = entry.find(ATOM + "content")
    if body is None:
        body = entry.find(ATOM + "summary")

    return {
        "title": entry.findtext(ATOM + "title"),
        "link": link,
        "date": entry.findtext(ATOM + "updated"),
        "body": body.text if body is not None else None,
        "media": media,
    }

Two details save the most time. Atom lives in a namespace, so every tag lookup needs the {http://www.w3.org/2005/Atom} prefix, which is why the RSS branch can use bare tag names and the Atom branch cannot. And the item body arrives as CDATA-wrapped HTML; a standard XML parser hands you the inner text already unwrapped, so body is an HTML string you may still want to strip of tags depending on what you are storing.

Handle media, dates and the fields that are not always there

Feeds are permissive, and every field above can be absent on a given item. The parser above uses findtext, which returns None rather than raising, and guards enclosure and content before touching them, because a podcast item has an <enclosure> and a text-only post does not.

Media specifically is the field most worth normalizing. RSS attaches it through <enclosure url="..." type="..." length="...">, Atom through <link rel="enclosure" href="...">, and returning both as a single media key means the rest of your code does not care which schema produced the item. If you need the media type as well, read the type attribute off the same node.

Dates are the other normalization worth doing at parse time rather than later. RSS pubDate is RFC 822 (Mon, 05 Aug 2026 12:00:00 +0000) and Atom updated is ISO 8601 (2026-08-05T12:00:00Z). Parsing both into a single datetime at the edge keeps every downstream comparison honest, the same normalize-at-the-edge approach used for cleaning scraped dates into UTC:

from email.utils import parsedate_to_datetime
from datetime import datetime


def to_datetime(value):
    if value is None:
        return None
    try:
        return parsedate_to_datetime(value)        # RSS: RFC 822
    except (TypeError, ValueError):
        return datetime.fromisoformat(value.replace("Z", "+00:00"))  # Atom: ISO 8601

Make the run reproducible and put it where it deploys

A feed scraper is usually a scheduled job on a server, which means it inherits the two things that make browser automation flaky in production: a non-deterministic identity and a proxy path you did not test.

Pin the identity. Passing seed= makes every fingerprint field come back identical run after run, so when a fetch starts getting challenged you can replay the exact same browser and tell a site change from a machine change. Set the proxy the job will actually use, and let the timezone follow the exit rather than pinning it by hand:

proxy = {
    "server": "socks5://gate.example.com:1080",
    "username": "user",
    "password": "pass",
}

with InvisiblePlaywright(seed=42, proxy=proxy) as browser:
    page = browser.new_page()
    page.goto("https://example.com/blog")

    href = page.get_attribute(FEED_SELECTOR, "href")
    feed_url = urljoin(page.url, href)

    page.goto(feed_url)
    items = parse_feed(page.content())

    for item in items:
        print(to_datetime(item["date"]), item["title"])

The proxy scheme and the SOCKS5 authentication details are their own topic, covered in Playwright SOCKS5 proxy with authentication; the important part for a scheduled feed job is that the endpoint carries an explicit port and that you confirm the exit inside the browser rather than assuming it. A minimal install for the container it runs in:

pip install invisible-playwright

The engine downloads and caches itself on first use, so the first run in a fresh container is slower than the rest.

Conclusion

Scraping feeds well is three separate jobs that people collapse into one. Discover the URL from the page’s <link rel="alternate"> rather than guessing a path. Fetch the XML through the same real browser that loaded the page, because the feed sits behind the same edge protection and a side client draws the challenge the browser does not. And parse defensively, branching on rss versus feed at the root and normalizing media and dates so the rest of your code never has to know which schema it got.

The feed being XML was never the hard part. Getting the XML, and getting it from a session that looks real enough to be served, is.

Short answers to the questions that lead here

Where do I find a site’s RSS feed URL? In the page head, as a <link rel="alternate" type="application/rss+xml"> (or application/atom+xml) element. Read its href and resolve it against the page URL, because it is often relative.

Why does fetching the feed with requests get blocked when the browser works? Because the feed endpoint usually sits behind the same edge protection as the HTML site, and a plain HTTP client is a different-looking visitor. Navigate to the feed in the browser and read page.content() instead.

How do I tell RSS from Atom? Read the root element. RSS has <rss> with a <channel> of <item> elements; Atom has <feed> with <entry> elements in the Atom namespace. Branch on that, do not assume.

Why is my parser returning empty results on some feeds? Almost always the schema. Code written for RSS <item>/<pubDate> finds nothing in an Atom feed of <entry>/<updated>, and Atom’s namespace means bare tag lookups miss.

How do I get the image or audio from a feed item? RSS uses <enclosure url="...">; Atom uses <link rel="enclosure" href="...">. Read both into one field so the rest of your code does not branch.

Why is the item body full of HTML tags? Because feed bodies are CDATA-wrapped HTML. A standard XML parser unwraps the CDATA for you and hands back the HTML string; strip the tags yourself if you only want text.

Sources

  • The RSS 2.0 specification and the Atom Syndication Format (RFC 4287) for the element and namespace differences described above, and RFC 822 for the date format RSS reuses.
  • The rel="alternate" link relation and Page.content() for the feed-discovery and fetch mechanics described above.
  • This project’s own testing method: discover from the page, fetch through the real browser past the same protection, and compare against a stock browser rather than reading a verdict, covered in how to test whether your browser is detected.

See also: how to capture XHR and API responses for reading a page’s JSON the same way, and how to scrape without getting blocked for the order to work in when the feed will not load at all.

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. The feed-behind-the-same-wall problem is one we hit ourselves before we wrote it down.

Back to top

MIT licensed. Every page here is written against the current source of the thing it describes, and several record something we got wrong first.

This site uses Just the Docs, a documentation theme for Jekyll.