Point sincewhen at the pages you care about — pricing pages, docs, changelogs, terms of service. It checks them on a schedule and answers with clean markdown diffs, over an API your tools can read and an MCP server your AI agents can call.
## Rate limits - Requests are limited to 100 per minute. + Requests are limited to 60 per minute. + Burst traffic above the limit returns 429.
A monitor that breaks usually goes quiet, and quiet reads as calm. A page that starts refusing our crawler, or a scoped watch whose selector stops matching after a redesign, looks exactly like a page that hasn't changed.
So silence is never allowed to answer for us. Every digest names the pages it could not check and why, and separately tells you how many changes are on file but fall outside your plan's history window. Expect that section — it is the feature, not an error.
Pages are watched once, not once per watcher, so a thousand people watching one page cost one fetch. The side effect is the useful part: add a URL someone already watches — with the same selector — and its change history is yours immediately, readable before your own first check has run.
Pages nobody has watched get their history reconstructed from public web archives instead, but only once the site has actually served us. We don't route around a crawler refusal to get it.
Add a URL. sincewhen fetches it on your plan's cadence, strips the boilerplate — nav, footers, scripts, image cache-busters — and keeps a normalized markdown snapshot.
Every check compares against the last snapshot. Real changes become diff events with added/removed line counts. When the only thing that moved is the page's clock, it's ignored — but a one-line price change on a huge page still counts, because that's the change you subscribed for.
Ask "what changed since Tuesday?" and get one markdown document across your whole watchlist — via REST, MCP, RSS, or webhooks. Made to be read by people and language models alike.
Every AI assistant has a knowledge cutoff. sincewhen is the patch stream: an agent can sign up, add watches, and pull diffs — no browser, no cookies, one round trip.
curl -X POST https://sincewhen.dev/api/auth/signup \
-H "Content-Type: application/json" \
-d '{"email":"you@example.com","password":"…","create_api_key":true}'
{
"mcpServers": {
"sincewhen": {
"type": "http",
"url": "https://sincewhen.dev/mcp",
"headers": { "Authorization": "Bearer sw_live_…" }
}
}
}
Tools: add_watch, list_watches, changes_since, get_diff, get_current_snapshot. Per-client setup on the connect page; full reference in the docs.
Works in clients that accept a bearer header — Claude Code, Cursor, VS Code. The Claude desktop, web and mobile apps connect over OAuth instead: paste https://sincewhen.dev/mcp as a custom connector, approve it, and there's no key to handle at all. Both paths are on the connect page.
$0
$5/month · or $48/year
Pro is built and works, but billing is not switched on, so there is currently no way to pay for it. The free plan is the whole product you can actually buy today, and it is free. More on that.