Gluecrawl is now a verified n8n node

Gluecrawl is now listed in the n8n integrations directory. The listing states that the integration is built and maintained by our partners at Gluecrawl and verified by n8n. The package is published as @gluecrawl/n8n-nodes-gluecrawl, currently at version 1.0.4, MIT licensed, by Deep Soup Labs.

Verification is the part that changes who can use it. n8n Cloud installs verified community nodes only, so before verification the node was reachable from self-hosted n8n and nowhere else. Now the same package installs on both, from the community nodes screen, without hosting anything yourself.

Nothing about the underlying product changed with the listing. Gluecrawl still takes a URL and a description of what you want, its mapper agents work out the selectors and pagination once, and every later run replays that configuration deterministically. The n8n node is a way to drive that from a workflow rather than from the dashboard.

The action node, Gluecrawl, has two resources. A Job is a saved URL plus what to extract from it, and its schedule. A Run is one execution of a job against the site, and the rows it extracted. Job operations cover Create, Get, Get Many and Delete. Run operations cover Start, Get, Get Many, Get Items and Download CSV, where Get Items emits one n8n item per scraped row so the next node receives a table rather than a blob of text.

The split matters for cost as much as for clarity. Creating a job runs the mapper and is charged upfront. Starting a run on a job that already exists replays the stored configuration and is billed for the pages it actually fetches. Anything recurring should create the job once and schedule the run.

The second node, Gluecrawl Trigger, works the other way around: it wakes a workflow when something finishes. It offers four events, Run Completed, Run Failed, Job Ready and Job Failed, with an optional job filter. Run Completed is the usual entry point, and the shape to prefer at volume, because the workflow costs no execution time while a scrape is running and starts only when there are rows to read.

AI newsletter reruns one Gluecrawl job per news site every Monday, keeps every article it has ever seen in an n8n data table, and asks an LLM to write a digest of only the stories that are new that week. Its setup branch creates the two tables and one job per site, each asking for the title, article URL, publication date and the article body. Requesting the body is deliberate: it cannot be read off a listing card, so it forces Gluecrawl to open each article. On Monday the archive acts as the dedupe key, so a story that sits on the front page for three weeks is written up once, and the whole week reaches the model in a single prompt so it can group related coverage instead of summarising story by story.

Local business leads picks a fresh business type and city each week with an LLM, mints a Gluecrawl job for that directory search, and scores every business it finds on how weak its web presence is. The picker reads the history of segments already swept, along with how many leads each returned and how many qualified, so it is choosing against results rather than against a list of names. The scoring never calls a model: a Code node fetches each business's own site and applies rules you can read and disagree with, where an unreachable or parked domain is the finding, a page that arrives as a script bundle is marked unreadable rather than scored as missing everything, and a domain that repeats across listings is treated as a chain. This is the one template that creates a job on every run, because a static directory has nothing new to say about a segment it already swept.

Property listing alerts reruns one Gluecrawl job per listing site every morning, keeps every home it has ever seen in a table, and emails only the ones that appeared since yesterday. It ships pointed at three portals in one city, asking each for the listing URL, address, price, beds, baths, square footage and ZIP code. A Code node puts the three portals into one shape and collapses the same house where it is listed on more than one of them, the table filter drops anything already seen, and the email is capped at the first fifteen new homes so that the opening edition does not run to hundreds and make every ordinary morning look empty. The setup notes ask you to sort each search by newest on the site itself before saving the URL, so page one is the page that changes.

The path is short, and each template documents the rest of its own setup on the canvas. Install the node from Settings, Community nodes, Install, using the package name. Create an API key in the Gluecrawl dashboard under Settings, API Key; keys are available on every paid plan, and there is one active key per account, so minting a new one revokes the previous key and any workflow still using it starts failing.

Then download the template JSON and import it with Workflows, Import from File. Templates ship with credentials unset by design, so open the Gluecrawl nodes and attach your credential before running anything. Read the sticky notes on the canvas: what to change before the first run is written next to the nodes that need it, including which nodes spend credits and which are safe to click twice.

Templates never arrive active. Run the scheduled branch by hand once, watch it end to end, and publish it only when the output is what you expected. A scrape spends credits, and the first execution of any of these workflows treats everything it sees as new.

The node does not try to be the whole workflow. It returns tabular JSON, which means the node after it can be a data table, a spreadsheet, a CRM, an email or another model, without a parsing step in between. All three templates end with a short list of what to hang off the last node, and in each case that change is a single node.

If you already run n8n, the fastest way to judge whether this is useful is to import one template, point its jobs at sources you care about, and let it run for two weeks. The first edition will be loud because nothing has been seen before. The second one is the one that tells you what changed.