Supporting a new Website

Feeds already supports a number of websites (see Supported Websites) but adding support for a new website doesn’t take too much time. All you need to do is write a so-called spider. A spider is a Python class that is used by Feeds to extract content from a website.

The feed generation pipeline looks like this:

  1. A spider extracts the content (e.g. an article) that should be part of the feed from a website. The spider also tells Feeds how the content should be cleaned up, e.g. which HTML elements should be removed.
  2. Feeds takes the content, cleans it up with the hints from the spider and some generic cleanup rules (e.g. <script> tags are always removed).
  3. Feeds writes an Atom feed for that site with the cleaned content to the file system.

A quick example

Writing a spider is easy! For simple websites it can be done in only about 30 lines of code.

Consider this example for a fictional website that hosts articles. When a new article is published, a link to it is added to an overview page. The idea is now to use that URL as a starting point for the spider and let the spider extract all the URLs to the articles. In the next step, the spider visits every article, extracts the article text and meta information (time, author) and creates a feed item out of it.

The following code shows how such a spider could look like for our example website:

import scrapy

from feeds.loaders import FeedEntryItemLoader
from feeds.spiders import FeedsSpider

class ExampleComSpider(FeedsSpider):
    name = ""
    start_urls = [""]
    feed_title = "Example Website"

    def parse(self, response):
        article_links = response.css(".article__link::attr(href)").extract()
        for link in article_links:
            yield scrapy.Request(response.urljoin(link), self._parse_article)

    def _parse_article(self, response):
        remove_elems = [".shareable-quote", ".share-bar"]
        il = FeedEntryItemLoader(
        il.add_value("link", response.url)
        il.add_css("title", "h1::text")
        il.add_css("author_name", "header .user-link__name::text")
        il.add_css("content_html", ".article-body")
        il.add_css("updated", ".article-date::text")
        return il.load_item()

First, the URL from the start_urls list is downloaded and the response is given to parse(). From there we extract the article links that should be scraped and yield scrapy.Request objects from the for loop. The callback method _parse_article() is executed once the download has finished. It extracts the article from the response HTML document and returns an item that will be placed into the feed automatically.

It’s enough to place the spider in the spiders folder. It doesn’t have to be registered somewhere for Feeds to pick it up. Now you can run it:

$ feeds crawl

The resulting feed can be found in output/

Reusing an existing feed

Often websites provide a feed but it’s not full text. In such cases you usually only want to augment the original feed with the full article.

Generic spider

For a lot of feeds (especially those from blogs) it is actually sufficient to use the Generic full-text extraction spider which can extract content from any website using heuristics (go to Generic full-text extraction for more on that).

Note that a lot of feeds (e.g. those generated by Wordpress) actually contain the full text but your feed reader chooses to show a summary instead. In such cases you can also use the Generic full-text extraction spider and add your feed URL to the fulltext_urls key in the config. This will create a full text feed from an existing feed without having to rely on heuristics.

Custom extraction

These spiders take an existing RSS feed and inline the article content while cleaning up the content (removing share buttons, etc.):

Paywalled content

If your website has a feed but some or all articles are behind a paywall or require to login to read, take a look at the following spiders:

Creating a feed from scratch

Some websites don’t offer any feed at all. In such cases we have to find an efficient way to detect new content and extract it.

Utilizing an API

Some use a REST API which we can use to fetch the content.

Utilizing the sitemap

Others provide a sitemap which we can parse:

Custom extraction

The last resort is to find a page that lists the newest articles and start scraping from there.

For paywalled content, take a look at:

Extraction rules

A great feed transports all the information from the original site but without the clutter. The reader should never have to leave their reader and go to the original site. The following rules help to reach that goal.

Unwanted content

Advertisement, share buttons/links, navigation elements and everything that is not part of the content is removed. The output should be similar to what Firefox Reader View (Readability) outputs, but more polished.


The HTML tags <figure> and <figcaption> are used for figures (if possible). Example:

<div><img src=""></img><div>
<figcaption>A very interesting image.</figcaption>

Credits for images are removed. Images are included in their highest resolution available.


If content is split in multiple pages, all pages are scraped.


Iframes are removed if they are unnecessary or untouched. Iframes are automatically replaced with a link to their source.

Updated field

Every feed item has an updated field. If the spider cannot provide such a field for an item because the original site doesn’t expose that information, Feeds will automatically use the timestamp when it saw the link of the item for the first time.

Not embeddable content

Sometimes external content like videos cannot be included in the feed because it needs JavaScript. In such cases the container of the external video is replaced with a note that says that the content is only available in the original content.

Regular expressions

Regular expressions are only used to replace content if using CSS selectors with replace_elems is not possible.


A feed item has categories taken from its original feed or from the site.


<h*> tags are used for headings (i. e. not generic tags like <p> or <div>). Headings start with <h2>. The title of the content is not part of the content and is removed.

Author name(s)

The name of all authors are added to the author_name field. The names are not part of the content and are removed.