Synthetic monitoring · MCP

You’ll set up monitoring
later. You won’t.

It’s forty minutes of forms, which is why it’s still on your list. Paste this instead. Your agent reads the repo and does it.

1 · connect

claude mcp add --transport http exit1 https://app.exit1.dev/mcp/v1 && claude mcp login exit1

Run in your terminal before starting claude. The login opens your browser to sign in.

2 · paste the prompt

Set up uptime monitoring for this project with Exit1.

1. Check whether the Exit1 tools (get_account, list_checks, create_check) are
   available to you. If they are, go to step 2.

   If they are not, do NOT try to connect from inside this session, and do not
   improvise with the raw API. Most AI tools load MCP servers only at startup
   and cannot open a browser sign-in mid-conversation, so a server added now
   stays unusable until I restart. Instead:
     - In Claude Code: run
         claude mcp add --transport http exit1 https://app.exit1.dev/mcp/v1
       then tell me to quit, run `claude mcp login exit1` in my terminal (my
       browser opens to sign in or sign up), and resume you with
       `claude --continue`. If my version has no `mcp login`, I
       authenticate via /mcp after restarting instead.
     - In other tools: tell me to add the HTTP MCP server
       https://app.exit1.dev/mcp/v1 the way my tool does it, restart it, and
       paste this prompt again.
   While you wait, do step 3 now and show me the checks you intend to create,
   so my first message after reconnecting can be "go".

2. Call get_account to see my plan limits, and list_checks to see what's
   already monitored.

3. Work out what's worth monitoring by reading this project. Don't ask me for
   URLs you can find yourself. Depending on the stack, look at deploy and
   infra config, environment files and examples, the README, DNS or domain
   config, container healthchecks, CI/CD workflows, and route or endpoint
   definitions. You're looking for:
     - the production site or app, and staging if there is one
     - health/status endpoints (these are the most valuable; assert on the
       response body, not just a 200, since a 200 with a dead database is
       still a 200)
     - public APIs and webhook receivers other systems depend on
     - scheduled jobs and workers (monitor these as heartbeats)
     - the apex domain, for SSL and registration expiry
   If this project has no deployed URL you can find, ask me for it.

4. Ask me two things in one go: which email address should receive alerts,
   and whether to add a Slack/Discord webhook. Then show me the checks you
   intend to create as one short list and create them when I confirm.

5. Send a test alert to every channel you set up and tell me to check it
   arrived. Don't leave a channel configured but untested.

Be fast and autonomous. Never put API keys, tokens or passwords into a check.
Prefer a few meaningful checks over many shallow ones.

Connect, then paste this.

exit1 · MCPworking

> Set up uptime monitoring for this project with Exit1.

Reading the repo…

✓vercel.jsonproduction domain

✓app/api/health/route.tshealth endpoint

✓.env.exampleAPI base URL

✓.github/workflows/nightly.ymlscheduled job

6 checks to create:

acme.comHTTP

api.acme.com/healthJSONPath $.status = "ok"

staging.acme.comHTTP

db.acme.com:5432TCP

acme.comSSL + domain expiry

nightly-billingheartbeat

Created 6 checks · alerts to you@acme.com and #ops

Test alert sent — confirm it arrived.

Get this.

Or sign up the normal way

50 monitors free. No card.

It’s synthetic monitoring

Boring term, precise one. Real requests hitting your endpoints from outside your infra, on a schedule. A 200 with a dead database is still a 200, so the checks assert on the response body with JSONPath. TCP ports, ICMP, SSL expiry and cron heartbeats as well.

Three things happen

Connect it

Pick your tool, paste one snippet, sign in from your browser. No API key. No account yet? You get one here.

Paste it

Claude Code, Cursor, Codex, VS Code, whatever you use. The agent reads the repo and proposes the checks.

Read what it built

It lists the checks, the alert channels, and the test alert it sent. Argue with it if it got something wrong.

Where this breaks. AI tools load MCP servers at startup and can’t open a browser sign-in mid-conversation. That’s why you connect before you paste. Pasted first anyway? The prompt tells the agent to stop, hand you the connect commands, and have you resume with claude --continue rather than guess.

Yes, it can write to your account

Here’s the blast radius. Every permission is on the consent screen before you approve, and you can revoke any client at app.exit1.dev/mcp.

It can

Create, update and pause checks
Read uptime, history and stats
Set who gets alert email
Wire up Slack, Discord, Teams, webhooks
Send a test alert

It can’t

Delete anything. checks:delete isn't granted unless a client asks for it.
Get around your plan limits. Every tool hits the same REST API, checked server-side.
Keep working after you revoke it. That kills every token for that client.
Act without your editor getting a chance to warn you. delete_check is flagged destructive.

Watches the stack you already run

After setup

Then stop opening dashboards

Same connection, read side. Ask in the editor you’re already sitting in. What’s down, what’s flaky, what broke last Tuesday.

Free is 50 monitors. Not a trial.

The MCP server has no plan requirement at all. Nothing to pay to find out whether this works.

50 monitors
5-minute checks
MCP access + 1 API key
300 alert emails / month
1 public status page
60-day history

Got users? Get 1-minute checks

Free checks every 5 minutes. That’s a 5-minute outage nobody told you about. Indie drops it to 1 minute, doubles you to 100 monitors, keeps 90 days of history. $3/mo on annual, after 7 days free.

Compare every plan

Go paste it

You’ve read enough.