# Why single-region uptime monitors wake you up at 3 AM — and how we built a two-region monitor with a read-only OAuth 2.1 MCP server

> Source: <https://dev.to/rares_corches_0967e149126/why-single-region-uptime-monitors-wake-you-up-at-3-am-and-how-we-built-a-two-region-monitor-with-1nmc>
> Published: 2026-11-22 14:08:56+00:00

**Key takeaway:** Beamfar's read-only MCP server at api.beamfar.com/mcp uses OAuth 2.1 with PKCE so AI coding assistants can query incidents and pre-incident diffs without users pasting API keys or granting write tokens.

If you have ever used a basic single-location uptime checker, you know the feeling: your phone buzzes at 03:14 with `CRITICAL: example.com is DOWN`, you jump out of bed, open the URL on your laptop, and it loads in 180 ms. Two minutes later: `RESOLVED`.

The internet between one probe datacenter and your origin has brief BGP reconvergences, packet loss spikes, and transient resolver timeouts every week. A single failed HTTP request from one datacenter is a **local network observation**, not proof that your website is down for users.

When designing [Beamfar](https://beamfar.com/?utm_source=devto&utm_medium=article&utm_campaign=two_region_mcp), we made two-region confirmation the core rule across every plan — including the €0 Free tier:

`verifying` instead of firing an alert immediately.
Knowing a site returned `502` or timed out is only half the story. Almost every production outage is preceded by a change:

Beamfar stores snapshots of DNS records, TLS certificates, final redirect URLs, and SEO headers (`noindex`, `robots.txt`, canonicals), and accepts optional deploy markers from our [GitHub Action](https://github.com/marketplace/actions/beamfar-deploy-marker) or official [WordPress plugin](https://wordpress.org/plugins/beamfar-monitoring/). When an incident opens, anything that changed in the previous hour is pinned right beside the incident timeline.

Once your monitoring data includes incidents, pre-incident diffs, SSL/domain expiry dates, and security findings, the next obvious workflow is asking your AI coding assistant inside the terminal or IDE:

```
Which sites had incidents in the last 7 days, and what changed right before each one?
```

Instead of making users copy-paste API keys into plaintext configs (or worse, granting write tokens), we exposed a **read-only MCP server** at `https://api.beamfar.com/mcp` using **OAuth 2.1 + PKCE** (registered in the Official MCP Registry as `com.beamfar/website-monitoring`):

```
# Claude Code
claude plugin marketplace add Beamfar/integrations
claude plugin install beamfar@beamfar

# Google Gemini CLI
gemini extensions install https://github.com/Beamfar/integrations

# OpenAI Codex CLI
codex mcp add beamfar --url https://api.beamfar.com/mcp
codex mcp login beamfar
```


