Whether something is up right now is a dashboard's job. What a conversation is better at is the follow-up — what happened to this endpoint over the last month, how long each outage lasted, is this the third time. Having incidents queryable by monitor is what makes that answerable in one turn. Creating monitors from a description is the other genuinely convenient bit: adding an endpoint mid-conversation is faster than any form.
An MCP server that connects an Uptime Agent monitoring account to an assistant. Uptime Agent watches websites and API endpoints and alerts when they fail; this puts the monitors and the incident record behind a conversation, and lets a new endpoint be added by saying so.
- Every monitoring endpoint listed, and one read in detail by id — `listMonitors`, `getMonitor`
- A new endpoint set up with a name, a URL, a tracking type such as http or ping, and a check frequency in seconds — `createMonitor`
- Detected downtime incidents listed, and one read in detail by id — `listIncidents`, `getIncident`
- All incidents for one specific endpoint, which is how you see whether a service has a history — `listIncidentsByMonitor`
- Public tracking created without authentication for a URL you want watched — `createAnonymousTracking`
An active Uptime Agent account and an API key: log in to the dashboard, go to Account then API Keys, create one with the permissions you need and copy it. Node.js 18 or higher. The quickest install is `npx uptime-agent-mcp setup`, which installs the server, configures it for Claude Desktop and prompts for the key; Smithery is an alternative. To do it by hand: clone, `npm install`, `npm run build`, and put `UPTIME_API_KEY` in a `.env` file, with `PORT` optional and defaulting to 3000. A Docker configuration is documented for clients that prefer it. Published as `uptime-agent-mcp` (0.0.10), MIT licensed.
One command plus a key — npx uptime-agent-mcp setup, then supply credentials