RSS Action

Fetches a feed and turns its items into payloads. It parses RSS, Atom, RDF and JSON Feed, remembers what it already emitted so a scheduled run only produces new items, and can go fetch the full article behind each link.

Setup

None. A public feed needs no credential — the node is an HTTP fetch with a parser and a memory.

Feed

The URL to fetch, and the format: Auto, XML or JSON Feed. Auto is right almost always — the parser has two paths, XML and JSON, and it detects which one to use from the body rather than trusting the content type.

ℹForcing JSON Feed is rarely what you want: a valid JSON feed is always detected. The setting earns its keep in the other direction — XML served with a JSON content type, which without the override would be handed to a JSON parser and fail.

How many, and filters

How many caps the items taken from the feed per run. Filters narrow them further — by title, by content, by date — so a busy feed can still drive a quiet flow.

Keywords, excluded words and authors ignore case and accents — envio finds “Envío”. A regex that does not compile, or a minimum date that is not a date (use 2026-10-01 or 2026-10-01T09:00:00Z), is refused when you save, instead of being skipped on every run.

Dedup across runs

On by default, and the reason the node is usable on a schedule: without it, every run re-emits every item still in the feed. With it, an item is emitted once and remembered, so a Scheduled Workflow pointing at a news feed produces new items only.

Run Test shows what the next run would emit, but does not mark anything as seen: you can test as many times as you like and the first real run still delivers those items.

Content enrichment

Many feeds carry a summary rather than the article. Enrichment fetches the linked page and pulls the full text, which is what makes the output worth handing to an AI Node or a Vector Store. It costs one HTTP request per item, so it pairs with a sensible How many.

Format and timeout

Format shapes each item before it leaves the node; Timeout bounds the fetch. A slow feed that blows the timeout fails the node rather than hanging the run.

Test RSS fetch

The panel fetches the feed on demand and shows what came back. It is the fastest way to find out that a feed needs enrichment, or that its dates are in a format the filter is not matching.

Mode — action or AI toolkit

The Mode selector decides who drives the node. In Single operation the node reads its feed every time the flow reaches it. AI toolkit turns it into four tools that an AI Node — or an MCP client, through one of your MCP servers — calls by name, choosing which one and with what arguments.

Tool	Arguments
get_feed_items	feed_url · limit? · since? · keywords? · exclude_keywords? · regex? · authors? · include_content? · full_articles? · format? · timeout_ms?
read_article	url
discover_feeds	url
list_feeds	—

get_feed_items returns the newest entries — title, link, date, author, categories and a short summary — with every option of Single operation as an argument.
read_article opens one page and returns its main text as Markdown, without menus or ads.
discover_feeds finds the feeds a website announces — or, when it announces none, the ones on the usual paths such as /feed.
list_feeds tells the model which feeds this node reads.

What you control

Which feeds the model can read. This node's feeds (the default) are the ones you list in the toolkit settings, one per line — or, when that list is empty, the Single operation's feed. Any other feed is refused before it is fetched. Any feed lets the model read whatever it names; your list then becomes its suggestions.
Nothing else. The model decides the rest from what the user asks — how many entries (20 unless it asks for more, at most 50), which words, authors or dates, and whether to open the full articles. The Filters, Output and Advanced tabs belong to Single operation: they are hidden in AI toolkit mode, and what was set there does not apply to the tools.
Characters per call sent to the model. 20,000 by default, anything from 1,000 to 200,000. Entries that don't fit are left out whole, and long articles are cut, with a note saying so.
Block these tool calls. None of the four starts blocked — they only read — but you can block any of them.

ℹDedup across runs does not apply in AI toolkit mode. Every call reads the feed as it is now, and the node's memory of what a flow already received is neither read nor changed — so asking twice gives the same answer twice. To get only what is new, the model passes since with a date.

Testing a tool

⚠Only a tool call picks the tool. In AI toolkit mode, only an AI Node, an MCP client or a voice agent chooses which tool the node runs. Run Test, a flow or a webhook run the operation saved on the node: an operation in their payload that names another tool is refused.

✦To try a tool, call it from an AI Node or from an MCP client. Its arguments look like this:

jsonA tool call

{ "operation": "get_feed_items", "keywords": "release", "since": "2026-09-01", "limit": 5 }

⚠A node in toolkit mode is not part of the pipeline. Events arriving through connections made before you switched are refused — the node only runs the tool it is asked for.

Gotchas

Without dedup, a scheduled run re-emits everything still in the feed, every time. It is on by default for that reason.
Enrichment is one extra request per item. On a feed of fifty items that is fifty fetches per run.
Feeds lie about their format. That is why detection reads the body instead of the content type — leave the format on Auto unless you have measured otherwise.
This is the RSS action, which fetches when the flow runs. To start a flow when a feed changes, use a trigger instead.