MCP server

Connect Alertum MCP

Ask Claude, Claude Code, or Cursor what's down, what got slower, or what happened overnight — grounded in your real monitors, incidents, and heartbeats. Read-only, team-scoped, nothing it can break.

$ claude mcp add --transport http alertum https://mcp.alertum.co/mcp$ claude mcp login alertum

Claude Code shown above. Claude Desktop and Cursor setup below.

MCP server

Ask your AI about your infrastructure

Alertum ships a hosted MCP server, so the AI assistant you already use can see your monitoring. Ask what's down, what got slower, or what happened overnight — and get answers grounded in your real checks, incidents, and heartbeats.

Ask in plain language

“What’s down right now?” “Why did checkout fail last night?” “Which certificates expire this month?” — your AI answers from live monitoring data, not from memory.

Your whole fleet, one conversation

Monitors, open incidents, heartbeats, and status pages are all reachable from the same chat — no dashboard tab-hopping while you investigate.

Safe to connect

The MCP server is read-only and team-scoped, with OAuth sign-in. Your AI can look at your monitoring — it can never pause a monitor, close an incident, or touch your API keys.

Works with the tools you already use

Claude, Claude Code, Cursor — anything that speaks the Model Context Protocol can connect in minutes. No SDK, no custom integration.

Full setup guide

Claude

Alertum MCP connected

Is anything down right now?

Yes — Payments API is down. It has been failing with HTTP 503 since 09:14, about 4 minutes now. An incident is already open and Maya from your on-call rotation was paged.

Your other 27 monitors are healthy — marketing site, auth service, checkout journey, and last night's backup heartbeat all look normal.

Was it slow before it went down?

It was. Response time climbed from ~300 ms to 1.8 s in the 20 minutes before the first failure — starting right after the 09:02 deploy of payments v2.4.1.

A gradual slowdown like that usually means the service was saturating, not crashing outright.

Who's on it, and what should I tell the team?

Maya acknowledged the page at 09:17 and is investigating. For the team: checkout is unaffected, the incident started at 09:14, and the likely trigger is the v2.4.1 deploy — worth having a rollback ready if Maya confirms it.

Claude setup

$ claude mcp add --transport http alertum https://mcp.alertum.co/mcp$ claude mcp login alertum

The seven tools

Every tool is a fixed, read-only lookup — nothing generic, nothing that writes.

alertum_list_monitorsList monitors with status; filter by status or search text

alertum_get_monitor_healthOne monitor's config plus uptime/latency statistics

alertum_list_open_incidentsOpen incidents with age and affected monitor, newest first

alertum_get_incidentOne incident's details — status, severity, root cause, timing

alertum_analyze_incidentPlain-language incident summary, key facts, and event timeline

alertum_list_heartbeatsHeartbeat (cron/job) monitors with interval and last ping

alertum_get_status_pageOne status page by id, or all pages when no id is given

Setup by client

Claude Code

Hosted over HTTP with OAuth sign-in — no API key to paste.

$ claude mcp add --transport http alertum https://mcp.alertum.co/mcp$ claude mcp login alertum

Claude Desktop

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json), then restart Claude Desktop.

{
  "mcpServers": {
    "alertum": {
      "command": "node",
      "args": ["/absolute/path/to/alertum-mcp/dist/index.js"],
      "env": {
        "ALERTUM_API_BASE_URL": "https://api.alertum.co",
        "ALERTUM_API_KEY": "sk-your-key"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json in your project, or ~/.cursor/mcp.json globally.

{
  "mcpServers": {
    "alertum": {
      "command": "node",
      "args": ["/absolute/path/to/alertum-mcp/dist/index.js"],
      "env": {
        "ALERTUM_API_BASE_URL": "https://api.alertum.co",
        "ALERTUM_API_KEY": "sk-your-key"
      }
    }
  }
}

API keys live in Alertum under Account → API Keys.

Security model

Read-only
Every tool maps to a fixed GET endpoint. No create, update, delete, pause, or admin tools — and no generic "call any URL" tool.
User-scoped via OAuth
Hosted mode uses OAuth-style MCP authentication. You sign in to Alertum during the connector flow; clients store MCP tokens, not your Alertum API key.
Team-scoped
Every response is scoped to your team. Resources outside your team return a clear not_found or forbidden error, never someone else's data.
No secrets in logs
Credentials come from the environment or the OAuth flow and are only sent to Alertum as request headers. Logging never includes credentials.

Frequently asked questions

Can my AI assistant change anything in Alertum through MCP?

No — the MCP server is strictly read-only. It can look at monitors, incidents, heartbeats, and status pages, but it can never pause a monitor, resolve an incident, or touch your API keys. For actions, use the Alertum Slack app or dashboard.

Do I need to paste an API key?

Not for the hosted connector — Claude opens an Alertum sign-in/consent page and the client receives MCP access and refresh tokens automatically. Local stdio setups (Claude Desktop config, Cursor) use an Alertum API key from Account → API Keys instead.

Which clients are supported?

Anything that speaks the Model Context Protocol — Claude Code and Claude.ai connect over hosted HTTP with OAuth; Claude Desktop and Cursor run the server locally via a small config file.

What happens if I ask about a monitor that isn't mine?

The Alertum API scopes every request to your team, so the tool returns a not_found or forbidden error rather than someone else's data.

Ready to connect your assistant?

Free plan available. No credit card required.

Get your API key View pricing