Infrastructure · Monitors
Know it is down before your customer does.
API uptime and performance scores in one dashboard. The alert reaches you by email, Discord or Telegram after two failures in a row, not after a network hiccup. A public status page on your domain answers your customers while you fix it.
Start for free, no card. The first monitor takes a minute: paste the URL, pick the interval, done. The dashboard is in Portuguese for now.
Why this exists
Today, the one who finds the outage is
THE CUSTOMER.
They call, send a message, open a ticket. By then the damage has happened twice: the service went down and someone outside had to tell you. Reversing that order is the whole product, and you do not have to build a clock, a history database and an alert channel to get it.
- The cron job that warns in a terminal nobody reads at 3 a.m.
- The alert that cries wolf: it fires so often it became ignored noise
- Support answering people one by one while the incident is still open
The whole path
Check, measure, alert, publish.
Four steps, and none of them is building infrastructure.
1 · Uptime
One request, three possible answers
Up, down, and up but slow, which is its own state. Waking someone at 3 a.m. for an API that answers correctly but slowly is how people stop trusting an alert.
kind: http url: https://api.acme.com/health interval: 30s · 1m · 5m · 10m · 30m · 1h → up answered as expected → degraded correct, above the threshold → down did not answer
2 · Performance
Real Lighthouse, in a real Chrome
Performance, best practices and SEO, with the browser opening the page. Each audit starts a full Chrome on its own machine and takes seconds. So audits leave the clock for a queue, and they never delay the pings.
kind: web url: https://acme.com interval: 1h · 6h · 12h · 24h → performance · best-practices · seo → serial queue of 1
3 · Alert
Two failures in a row, not one
An incident opens after N bad checks in a row (default 2) and closes on the first good one. A single failure, like the network timeout everyone gets, does not become a notification. That is what keeps the alert meaningful.
check 1 down → nothing yet
check 2 down → incident OPENED
↓
email
Discord
Telegram
webhook
check 3 up → incident CLOSED4 · Public page
Status on your customer-facing domain, with your look
Color, font, theme and footer are yours. So is the address. You point a CNAME and the certificate is requested automatically. During an incident, the page answers for you while support is still reading the alert.
status.acme.com CNAME → cname.monitors.myinfrastructure.click ✓ certificate requested automatically ✓ HTML ready on the first byte ✓ 90 days of history before DNS: /s/<slug>
Model Context Protocol
Your agent investigates the outage. Without opening the dashboard.
Looking at the chart is the easy part. What was missing is answering what comes after it: what happened, when it started, and whether it is back to normal. There are 18 tools, 9 that read and 9 that write, on revision 2026-07-28, the stateless one. The address covers the whole account: the key already covers all your monitors, so there is no id in the path.
What the server does NOT offer
- Deleting a monitor or status page does not exist as a tool. This matters more here than in other products: deleting a monitor also deletes the history that would prove the incident.
- Adding an alert channel or custom domain stays in the dashboard, where a person confirms it. An alert target chosen by an agent is an exfiltration surface: alerts would start going to an address nobody chose.
- And the plan still applies creating a monitor or a page goes through the same limit as the API. Over the plan, the call fails and says how much fits. The agent reports instead of retrying.
Authenticates both ways with your mk_ key or with OAuth 2.1 through Infrastructure Auth. Both reach the same account.
The yardstick
The five criteria that decide the choice
The same table on every Infrastructure product page. Two rows here are limitations, written in the same tone as the others. It exists so you can quickly decide whether this fits you.
What gets checked
Two families, one dashboard
API uptime by HTTP request (from 30s) and Lighthouse audits in a real Chrome (from 1h). Their costs differ by orders of magnitude, about 1 KB versus a full browser, so their minimum intervals differ too.
Where the check runs from
One region, no replica
One point on the Cloudflare network per check. A network failure between us and you shows up as your outage. If you need to tell a global incident from a bad route, you need probes in several places, and that does not exist here today.
When the alert fires
After 2 failures in a row
Configurable. Closes on the first good check. Sending never holds the clock: short timeout, no retry, and the incident is recorded either way. The dashboard is the source of truth; the alert is the convenience.
Where the alert goes
Email, Discord, Telegram and webhook
The email list accepts the whole on-call roster pasted at once: commas, semicolons or a spreadsheet column. No SMS, no phone calls, no on-call rotation: everyone on the list gets it together. Internal network targets (localhost, 169.254.169.254, private ranges) are rejected. The request leaves our infrastructure, and accepting them would turn the service into a network reader for whoever registers the channel.
Agent integration
MCP, 18 tools
MCP server on revision 2026-07-28, authenticating with OAuth 2.1. You approve in the browser, with no key in a file. Nine tools read and nine write. None delete, neither monitors nor status pages. Adding alert channels and custom domains also stays out, in the dashboard.
If any of these points does not match your case, write to us. We will check and fix the row.
Pricing
3, 9 or 199 monitors
FAQ
The objections, in the order they come up
Found a limit that is not written here?
Talk directly to the people who wrote the code. If you found where it breaks, we want to publish it before the next visitor finds it alone.
Paste the URL, pick the interval. It takes a minute.
No card. If your service goes down tonight, the alert tells you, not your customer.