RSS Feed Reader: RSS, Atom & JSON Feeds, Keyword Alerts

Get every item from any RSS, Atom or JSON feed as clean rows, or schedule it and get only the new items since the last run, filtered by your keywords, with one-line alerts ready for Slack.

• Any feed, or just the website. RSS 2.0, RSS 1.0 (RDF), Atom and JSON Feed. Paste a website instead and the Actor finds the feed it links to, or (for a bare domain) tries /feed, /rss.xml, /feed.xml, /atom.xml and /index.xml.
• Monitoring built in. Turn on Only new items since the last run, schedule it, and you pay only for items you have not received before. Feeds drop old items all the time, so nothing is ever reported as "removed" and an item that briefly disappears does not come back as new.
• Keyword alerts that make sense. Whole-word, any-case matching on title, description and categories: AI finds "AI-powered" but not "said". Add exclusions, a "published within N days" filter, and pay only for the items that match.
• One story, one row. Stable ids even for feeds that change their item ids (the BBC adds a counter such as #0 or #4 to the same story), so you are not alerted or charged twice. A site that is down for maintenance, or a feed that moves, never makes old items come back as new.
• Alerts out of the box. Every row has a one-line summary, and the run's status message lists the first new items, so the built-in Slack integration posts "New in BBC News: …" with no extra tools.

It reads only the feeds themselves: one request per feed (a website adds its home page and the usual feed locations). No browser and no proxies, so runs take seconds and cost little.

Sample output

{
  "status": "new",
  "summary": "New in BBC News: Google value plunges by $500bn as AI fears rattle Wall Street https://www.bbc.co.uk/news/articles/cm9qzdqpxy48o?at_medium=RSS&at_campaign=rss",
  "feedUrl": "https://feeds.bbci.co.uk/news/technology/rss.xml",
  "feedTitle": "BBC News",
  "itemId": "https://www.bbc.co.uk/news/articles/cm9qzdqpxy48o",
  "title": "Google value plunges by $500bn as AI fears rattle Wall Street",
  "url": "https://www.bbc.co.uk/news/articles/cm9qzdqpxy48o?at_medium=RSS&at_campaign=rss",
  "publisher": null,
  "publisherUrl": null,
  "publishedAt": "2026-10-29T21:52:13Z",
  "updatedAt": null,
  "description": "Google spooked investors so badly that its stock hit a six-month low, after it revealed its AI search summaries were affecting ad revenue.",
  "categories": [],
  "imageUrl": "https://ichef.bbci.co.uk/ace/standard/240/cpsprodpb/3304/live/1c4d6e00-d3d4-11f1-b441-2b33b063138f.jpg",
  "firstSeenAt": "2026-10-29T22:30:00Z",
  "scrapedAt": "2026-10-29T22:30:00Z"
}

A real row from the BBC Technology feed, monitored with the keyword AI. Dates are converted to UTC ISO 8601 whatever the feed uses (Thu, 29 Oct 2026 21:52:13 GMT, +0200, CEST, ISO 8601).

Who uses it

• PR and communications teams: media monitoring of news sites and trade press for your brand, clients or competitors, as Slack alerts.
• Marketers and content teams: every new post on competitor and industry blogs, in one sheet, every morning.
• Analysts and investors: press releases, regulators' and central banks' news feeds, filtered by company names or topics.
• Developers and newsletter builders: clean, de-duplicated JSON from hundreds of feeds for an aggregator, a search index or an LLM pipeline.
• AI agents: a narrow, predictable tool for "what is new on these sites about X?"; works through the Apify MCP server.

How to use it

1. Paste feeds or websites into Feed URLs or websites, one per line, for example https://feeds.bbci.co.uk/news/technology/rss.xml or example.com/blog. For news coverage of a company or topic, follow the news sites' own feeds and add your company or topic as keywords.
2. Optionally add Only items mentioning (keywords), Skip items mentioning and Published within the last N days.
3. Click Start. Download the results as JSON, CSV or Excel, or read them through the API.

To monitor: turn on Only new items since the last run, save the input as a task and add a schedule (hourly or daily). The first run returns every current item marked new (or, with Start quietly, just records them for the price of the feed check); after that you get only new items. Connect the task to Slack, email, Google Sheets, Make, Zapier or n8n to get alerts.

Input examples

All items from two feeds:

{ "feeds": ["https://feeds.bbci.co.uk/news/technology/rss.xml", "https://feeds.bbci.co.uk/news/science_and_environment/rss.xml"] }

Hourly keyword alerts from news sites and a competitor's blog, starting quietly:

{
  "feeds": ["https://feeds.bbci.co.uk/news/business/rss.xml", "competitor.com/blog"],
  "keywords": ["acquisition", "Acme Corp", "regulat*"],
  "excludeKeywords": ["podcast", "sponsored"],
  "onlyNew": true,
  "skipExistingOnFirstRun": true
}

Inputs

Input	What it does	Default
feeds	Feed URLs (RSS, RDF, Atom, JSON Feed) or websites, one per line	required
keywords	Keep only items whose title, description or categories mention any of these (whole words, any case, * for any ending)	all items
excludeKeywords	Drop items that mention any of these	none
publishedWithinDays	Keep only items published in the last N days (0 = any date)	0
onlyNew	Monitoring: only items not received before	off
skipExistingOnFirstRun	Monitoring: the first run records today's items without returning them	off
reportUpdates	Monitoring: also return known items whose title or updated date changed (changed, with the previous values)	off
maxItemsPerFeed	Newest N matching items kept per feed (changing it keeps the monitoring memory)	200
includeContent	Add the summary as plain text (about 1,000 characters)	on
stateStoreName, maxConcurrency	Advanced: monitoring memory name, feeds read in parallel	defaults work

Which inputs work

You enter	What happens
A feed URL, with or without .xml, for example https://feeds.bbci.co.uk/news/rss.xml	That feed is read
A bare domain, for example example.com	The feed its home page links to (<link rel="alternate">; comment feeds only if nothing else is linked); if the page links none, /feed, /rss.xml, /feed.xml, /atom.xml, /index.xml
A page, for example https://example.com/blog	The feed the page links to
A news.google.com link	Refused with a free invalid_input row: Google's terms allow its news feeds for personal, non-commercial use only

Inputs that cannot be used produce a free row with status: invalid_input, not_found or error and a plain-English error. The Actor never guesses around trouble: a home page that answers with an error (403, 503), or a feed URL that answers with a web page (a maintenance page), is an error row, not a reason to read some other feed. When monitoring, a website whose page links no feed on one run (a maintenance page that still answers 200) is read through the feed it linked before, never a guessed one.

The same feed entered twice, or a website together with its own feed, is read and charged once: a feed URL wins over a website, then the earlier input. The other input gets no row (RUN_SUMMARY lists it as duplicate) and is not counted as a problem. Both share one monitoring memory, because memory is kept per feed: if the feed URL fails on one run and the website reads the same feed, nothing is sent again.

Output

One row per item: status, summary, feedUrl, feedTitle, itemId, title, url, publisher and publisherUrl (for feeds that name a source, such as news aggregators), publishedAt, updatedAt, description, categories, imageUrl, firstSeenAt (monitoring) and scrapedAt. status is current in a normal run and new (or changed, with previousTitle and previousUpdatedAt) when monitoring. Items come newest first. Author names and emails are never included. Inputs that produced no items get a free row with status invalid_input, not_found, error or skipped (not checked because your spending limit was reached or the run stopped early; one row for all of them, see the FAQ); an input that leads to a feed already read for another input gets no row.

A RUN_SUMMARY record in the run's key-value store lists every input with the feed that was read, its format, how it was found, and how many items it had and matched.

Get alerts in Slack, Google Drive, Make, Zapier or n8n

1. Turn on monitoring, click Save as a new task, and add a schedule to the task.
2. On the task's Integrations tab, pick Slack, connect your workspace, choose the Run succeeded event and a channel, and paste this as the message:

*{{resource.statusMessage}}*
<https://console.apify.com/storage/datasets/{{resource.defaultDatasetId}}|Open all items>

The status message reads like Checked 3 feed(s), saved 2 item row(s) (2 new). New in BBC News: Google value plunges by $500bn as AI fears rattle Wall Street https://www.bbc.co.uk/news/articles/… so the alert itself has the headline and the link.

• Google Drive: the Upload results to GDrive integration saves each run's rows as CSV or JSON.
• Make, Zapier, n8n: use the Apify app's "Actor run finished" trigger and map the summary, title and url fields into a message, a sheet row or a CRM note.
• Webhooks: send the run to your own endpoint and read the dataset through the API.

Pricing

Pay per event, no platform usage charges on top:

Event	Price
Item returned	$0.50 per 1,000 items
Feed checked	$1.00 per 1,000 feeds

Rows for inputs that could not be used are free. A feed check is charged after its rows are delivered, so if your spending limit stops the run before any of a feed's rows go out, that check is free.

• One-off export: 50 news and blog feeds with about 40 items each (2,000 items) cost about $1.05: $1.00 for the items and $0.05 for the checks.
• Hourly keyword alerts: 25 feeds checked every hour is 600 checks a day ($0.60); with about 20 matching new items a day ($0.01) that is about $0.61 a day, about $18 a month. Checked every 6 hours instead: about $3.30 a month. You never pay for items you already received (changing the max items per feed, or a site's maintenance page, does not re-send them) or for items that do not match your keywords.

Set a maximum cost per run in the run options and the Actor stops cleanly when it reaches it: a feed is read only when the limit still covers its check and one item ($0.0015), feeds it could not check share one free skipped row, and when monitoring, new items it could not deliver come on the next run. Rows it delivered are never charged twice.

FAQ

Is this allowed? Feeds are published so that readers and other software can follow a site; the Actor reads only those feeds (plus, for a website input, the home page and the usual feed locations), one request at a time per feed, with no login. Author names and emails are left out. Feed content still belongs to its publisher: follow each publisher's terms for reuse (the BBC, for example, publishes its own terms for its feeds) and the terms of any service you send results to.

Why are items never reported as removed? A feed shows only the latest items, so items leave it all the time; that is not news. The Actor remembers each item until 30 days after it was last in the feed, so an item that drops out and comes back (or a feed that is briefly empty) is not reported again, however long the item has been listed.

The BBC (or another site) lists the same story twice. Do I pay twice? No. When a feed's item id is the article's address plus a counter while its link is that address with tracking parameters (the BBC uses #0, #1, … and the same story can appear as #0 and #4 in one feed), the counter is ignored, so one story is one row. Posts whose ids are anchors on the very page they link to (a live blog's updates) keep their own ids, even when every link carries ?utm_source=rss. Items that share one id but link to different pages (deal-of-the-day and chart feeds reuse one) are told apart: the item first seen under that id keeps it, and each of the others gets the id plus its link. So an item you already received does not come again when it leaves and returns or is retitled, and a new item that reuses an old id is never missed. One trade-off: a page that moves to a new address while keeping its id is sent once more, because it looks exactly like a new item reusing the id (tracking tags and session ids in links are ignored). Ids made from links ignore utm_ tracking parameters and http versus https.

I follow several feeds of one site. Will a story in two of them come twice? Yes, once per feed: each feed has its own memory, so a story the BBC lists in both News and Technology arrives from both. Follow the narrower feeds, or use keywords, if that is too noisy.

How do keywords match? Whole words or phrases in the title, description and categories, any case. AI matches "AI" and "AI-powered" but not "said" or "MAIN"; regulat* matches "regulation" and "regulators"; machine learning matches across line breaks. A straight apostrophe matches curly ones (McDonald's finds "McDonald’s"). When a feed's description holds only an image, the full content is searched instead. Keywords in languages written without spaces (Chinese, Japanese) match anywhere in the text. Changing keywords, exclusions or the date window starts a fresh monitoring memory, so a new filter does not hide items you have not received yet (turn on Start quietly to skip that fresh start's first batch). Changing the max items per feed keeps the memory: lowering it sends nothing again, and after a raise the older items it brings in (dated before the oldest item of the previous run) are recorded without being sent, while items published since the previous run still come.

A website has several feeds. Which one is read? The first feed its page links to (comment feeds only when the page links nothing else). To choose, paste that feed's URL. The feedUrl column and RUN_SUMMARY show which feed was used. If the home page or the linked feed is down, the run reports an error for that input rather than quietly switching to another feed, so a monitor never mistakes a different feed for new items. When monitoring, the Actor remembers which feed each website led to: if the page links no feed on a run (a maintenance page), that feed is read again. If the site starts linking a different feed, its first feed is missing for a run, or a feed URL redirects to a new address, the memory comes along (and back again), so items with the same ids are not sent again. Items with ids you have not received are new, even with Start quietly on: a site that switches to a genuinely different feed (say from its news feed to its podcast) sends that feed's current items once.

What about broken feeds? Common mistakes that feed readers forgive are repaired: HTML entities such as  , unescaped &, control characters, Windows-1252 characters in a UTF-8 or Latin-1 feed (curly quotes, é), namespace prefixes used without being declared, a byte-order mark or a PHP warning before the XML, gzip, UTF-16 and Shift_JIS or GB2312 encodings. One item with a malformed link never costs the rest of the feed. Dates with +02:00 offsets, 9:52 PM times and zone names such as CEST or SAST are converted correctly; a zone name the Actor does not know leaves the date as written rather than guessing UTC. A feed that stamps every item with the same date (its build time) has those dates ignored, so a rebuild is never reported as updates. Documents that declare entities are refused (that blocks entity-expansion attacks), and a feed cut off mid-download is an error, never a short list.

Why is the first monitoring run bigger? Nothing has been seen yet, so every current item counts as new. Turn on Start quietly to skip that. Each saved task, and each combination of filters, keeps its own memory, and normal (non-monitoring) runs never touch it.

Can it monitor Google News? No. Google's terms allow its news feeds for personal, non-commercial use only, so this paid Actor does not read them: a news.google.com link gets a free invalid_input row, and a page that links or redirects to Google News is not followed there. Follow the news sites' own feeds instead (most publishers have them) and use keywords to pick out your company or topic. Feeds that name a source (<source> in RSS, common in aggregators) fill publisher and publisherUrl.

Why did my run stop early? Four limits can end a run before every input is checked, and none of them charges you for what was not checked: If one Apify account's runs keep producing nothing to charge (for example a list of dead sites run again and again), new runs from that account stop early for up to 24 hours, with a free row that says so; normal use never gets near this.

• Your spending limit. A feed is read only when the run's maximum cost still covers one feed check and one item ($0.0015); the rest share one free skipped row and come on the next run.
• Many failures in a row. After 50 feeds in a row fail (not found, errors, timeouts), the list or the sites are probably down, so the run stops and the rest share one free skipped row. A site that fails 10 times in a row is not called again in that run.
• A safety limit. A run that does a lot of work without returning anything you pay for (for example thousands of dead feeds, or keywords that match nothing in huge feeds) stops after a few minutes of such work, with one free skipped row. Runs that return items are never stopped by it.
• Too many inputs. A run checks at most 1,000 feeds or websites (the first ones entered); split longer lists across several tasks. Keyword and exclusion lists take up to 100 entries each.

The status message and RUN_SUMMARY (stoppedEarly, notChecked, skippedBySpendingLimit) say which one it was.

Can it read feeds on my own network or server? Only public ones. Addresses on private networks and the local machine (localhost, 127.0.0.1 or 127.1, 10.x.x.x, 192.168.x.x, cloud metadata such as 169.254.169.254, and names ending in .local or .internal) are refused with a free invalid_input row, whatever the spelling. The same check runs after the name is looked up and on every redirect, so a public URL that redirects to a private address gets a free error row instead. Publish the feed on a public address to monitor it.

Can I use it from code or an AI agent? Yes: through the Apify API, the Python and JavaScript clients, or the Apify MCP server. Inputs and outputs are stable and documented above.

Support

Open an issue on the Issues tab with the feed or website you tried; I aim to answer within a day.