Product Tour Pricing Docs Changelog About Sign in Start free
NEW Ask Mercury why it broke →

Uptime monitoring that
doesn't cry wolf.

All three probes, on three continents, have to agree before anyone gets paged. Single-region flaps are logged, not escalated. Status pages and on-call included.

Real probes in Amsterdam, Virginia and Singapore. Or skip it and start free.

25 monitors free forever, commercial use included. New accounts start on a 14-day Team trial — no card required.

Replay · from our own pager repose.sh — landing · 24 September · UTC
  1. 15:49:16AmsterdamHTTP 200 · 108 ms
  2. 15:49:22SingaporeHTTP 200 · 77 ms
  3. 15:49:24VirginiaHTTP 200 · 134 ms
  4. 15:50:18AmsterdamHTTP 200 · 111 ms
  5. 15:50:27VirginiaHTTP 200 · 119 ms
  6. 15:50:29Singaporefetch failed
  7. 15:50:33Engine1/3 regions failing — below consensus threshold of 3. Logged as anomaly, no page.
  8. 15:51:19AmsterdamHTTP 200 · 119 ms
  9. 15:51:25SingaporeHTTP 200 · 96 ms
  10. 15:51:30VirginiaHTTP 200 · 109 ms
  11. 15:51:33Engine0/3 regions failing. Nothing to do.
Nobody was paged. Singapore couldn't reach it; Amsterdam and Virginia could. 60 seconds later all 3 agreed it was fine. A monitor that pages on the first failed check would have woken someone for this.

Not a mock-up. These are the rows our production engine wrote, on one of our own monitors. See the working →

Last 30 days · live from our own engine

The quiet is measurable.

325,534
probes run
From Amsterdam, Virginia and Singapore — real machines, not region labels.
110,495
decisions logged
Every page and every non-page, with the reason it went that way. Never deleted.
89
pages held back
99% of everything that could have woken somebody. It didn't.

Counted live from production, including our own test workspaces — which is why the numbers aren't round. See the working.

The alert engine

One region disagreeing
is not an outage.

Most tools page on the first failed check. That is why your phone has gone off for a datacenter that fixed itself in forty seconds. We wait until the failure is real.

eu-west us-east ap-southeast
Held 1 of 3

Logged as an anomaly with the full decision trail. Visible in your dashboard tomorrow morning. Nobody is woken.

eu-west us-east ap-southeast
Paged 3 of 3

Three probes, three continents, inside thirty seconds. Mercury has already written the first investigation. Now it's worth waking for.

And when the cause is an upstream provider having a bad day, we tag it instead of blaming you — we mirror 14 status pages to check.

What comes with it

Everything the others
charge extra for.

Status pages, no subscriber tax
Unlimited viewers on every plan. Email subscribers: 25 on Free, unlimited from the $10 plan up — never priced per subscriber. Statuspage charges $99/mo for a thousand. Bring your own domain and colours.
On-call, bundled
Rotations, layered escalation, vacation cover, and handoffs at 09:00 your time — not UTC midnight. PagerDuty starts at $19 per user.
Mercury investigates
The moment an incident opens you get the probe evidence, the likely cause, and what to check first. Then you can ask it follow-up questions.
An audit trail you can't edit
Every alert decision is recorded with its reasoning — including the ones where we chose not to page you. Nothing is ever deleted.
Built for agents
A native MCP server and a REST API. Point Claude or Cursor at your workspace and describe what you run; it configures the monitors.
Phone alerts that arrive
Install Repose as an app and get push notifications when consensus fires. Only for pages that matter — recoveries stay silent.
Before you ask

The awkward questions.

Is the free tier a trial?

No. 25 monitors, forever, commercial use included. New accounts also get 14 days of the Team plan to feel the difference — no card. When it ends, checks slow to every 3 minutes and nothing switches off.

What if Repose goes down?

We watch ourselves with Repose and the status page is public. If one of our probe regions fails, the consensus bar drops to the regions still reporting and every decision says so — our outage must never hide yours.

I already use UptimeRobot.

Then you already know what a 3am false alarm costs. Point Repose at the same endpoints for a fortnight and compare what each one woke you for. That comparison is the whole pitch.

Will you page me less than I need?

Consensus is per-monitor. Set it to 1 and we behave like everyone else. The default is 3 because that's what the evidence supports — but it's your call, per service.

Your first monitor,
in under two minutes.

No card. Commercial use fine. We'll only wake you when all three probes, on three continents, agree — and Mercury will already know why.