Have an agent build it. Copy the brief below into Claude Code, Cursor, or any coding agent with access to your project. It states the calls, the questions worth asking you first, and the mistakes to avoid.
Agent brief
Agent brief
The pipeline
1. Pick what to watch
There are three kinds of target, and picking the right one is most of the job.- Pages when you care about what a page says: a pricing page, a changelog, terms of service. You get a diff of the text.
- A site when you care about what gets published: a blog, a careers section, a product catalogue. You get each new page with its title and opening lines, and pages that go away.
- A search when you care about what others say: coverage, mentions, reviews. You get results that were never there before.
2. Create the monitor
3. Receive the change
Each change arrives as amonitor.change call. Verify the signature first, as described in Verifying a webhook, then use the status:
changed: show thediff, andfieldChangeswhen a price or availability moved.new: show thetitleandexcerpt, with a link to theurl.removed: the page was missing from two checks in a row.error: the page could not be read three checks in a row.
monitor.check.completed arrives once a check ends, with its counts. Use it for a summary, or to notice a watch that went quiet.
Making the alert worth reading
An alert saying “rival.io/pricing changed” is a task, not information. Every change already carries the diff or the new page’s opening lines, so put that in the message. Then cut what you do not care about, before it reaches anyone.- A goal in plain words, like
a plan gets cheaperorsenior roles in Berlin, alerts only on changes that match it and keeps the rest in the history with the reason it was held back. 1 credit per change checked. selectorcompares only part of a page, such as#pricing, so a new footer link or a rotating testimonial never fires.triggerOnalerts only when a changed line contains a word you name, likepriceor/sold out/i.ignoreleaves out lines that change without meaning anything.
Knowing the watch still works
A watch that fails silently is worse than none, because you believe you are covered. Monitors report their own health. A page that cannot be read three checks in a row sends anerror alert instead of staying quiet, and a monitor whose checks fail ten times in a row pauses itself with pausedReason: "failing" and says so in its last monitor.check.completed call. Every check, including the quiet ones, is in GET /v1/web/monitors/{id}/checks, so a month without changes can be told apart from a month without checks.
A site target lists the site again on every check, so a page at a URL you never knew existed still gets noticed.
What it costs
- Six pages checked daily: 6 credits a day, about 180 a month.
- A blog checked daily: 1 credit a day, plus 1 for each new post read.
- A goal adds 1 credit for each change it checks, not for every check, so on a quiet page it costs almost nothing.
What to watch for
Daily is almost always enough. Nobody’s pricing page changes hourly, and the shortest schedules are for status pages and stock, not for terms of service. A monitor with a search target checks at most once an hour. Start without a goal for a week, read what the monitor reports, then add a goal or aselector for the noise you actually saw. Guessing the noise in advance usually filters out the change you wanted.