Website Change Monitor

Everybody means to check the competitor's pricing page and the vendor docs their billing depends on, and nobody does it twice. Firecrawl watches the addresses; this agent is the part that reads the page after a check fires and writes the sentence a human needs, which is what the wording was, what it is now, and who here should care. Cosmetic churn is thrown out before Slack sees it, and that is the difference between an alert you keep switched on and one you mute in week two.

What people use it for

• Competitor pricing, the day it moves - Sales find out a rival dropped a tier from a lost deal six weeks later. Quoting the old and new price side by side the same morning gives them something to answer with.
• Vendor deprecations before they break you - The docs page for an API you depend on grows a removal date months ahead of the outage. Reading it that morning is the whole gap between planned work and an incident.
• Cosmetic churn stays out of Slack - Careers pages, testimonials and build hashes move nightly and mean nothing. Naming that churn once teaches the agent to end those runs in silence.
• Proof of the old wording - Every note keeps the previous text sitting next to the new. Months later that Slack thread is the only proof a policy, a limit or a public claim was ever phrased differently.

Before you fork

What has to be set up on the Firecrawl side?

A Firecrawl key and the monitors themselves, created there with the addresses and the check frequency you want. This workflow reads the checks and scrapes the flagged page; it cannot add a monitor, so your watch list only grows when you decide it does. Slack is the other connection, and Test Run produces a finished note from a staged check before anything runs live.

Is it doing anything to the sites it watches?

It fetches public pages, one address at a time, exactly as a browser would. There is no logging in, no form filling, no clicking into a signup or a checkout, and no crawl of the wider domain, all of it stated in its instructions and none of it wired into the workflow. Pages that require an account are simply reported as unreadable.

A page we watch changes every single night. What then?

Each monitor event costs one agent run whether or not anything is said, so a nightly churner is a real cost with no output. Describe that churn in the watched pages notes and it will end those runs silently, and if the page is genuinely worthless, either slow its check frequency in Firecrawl or remove the monitor there. The agent will never make that call for you.