Skip to content
upcertain

Blog

Ask Claude what's down: connecting your uptime monitors over MCP

During an incident, the first questions are always the same: what is down, since when, what else is affected, has this happened before? With Upcertain connected over MCP, you can ask Claude instead of clicking through dashboards.

Updated 25 October 2026 · By the Upcertain team

What MCP gives Claude

The Model Context Protocol lets an AI client call tools on a server. Upcertain's MCP server at https://upcertain.com/api/mcp offers tools to list monitors, read a monitor's uptime over 24 hours, 7, 30 or 90 days, list and read incidents with their activity, and list status pages. With read-write access it can also create, pause and resume monitors and acknowledge incidents.

Connect Claude

Add https://upcertain.com/api/mcp as a custom connector in Claude, or as an HTTP server in Claude Code, and sign in when Upcertain opens in your browser. You choose the workspace and read-only or read-write access there. The steps for each client are in the docs for Claude, Claude Code and Cursor. MCP is included on the Team and Business plans.

A useful habit is to start an incident conversation with the broad question and narrow from there. Claude calls the list tools first, then fetches the incident or monitor you ask about, so you see the same records your team sees in the app.

Questions that work well

During an incident

  • "What's down right now, and since when?"
  • "Did anything else fail in the ten minutes before checkout went down?"
  • "Has the payments API had incidents like this before? Show the last five with causes."

For reports

  • "Make a table of 30-day uptime for every monitor, worst first, with the number of incidents."
  • "Summarise this week's incidents for a status update to customers."
  • "Which monitors have had the most incidents in the last 90 days?"

For changes (read-write)

  • "Add HTTP monitors for these five URLs every minute, expecting 200."
  • "Pause the staging monitors; I'll ask you to resume them after the migration."
  • "Acknowledge the open incident on auth-service."

Safeguards we built in

  • Read-only by default. Every new connection starts read-only. Read-write is offered only to people whose role can make changes.
  • One person, one connection. Each sign-in is its own connection, listed under Settings, API & MCP, Connected apps, and can be disconnected by its owner or by an account owner or admin.
  • Follows the membership. Leave the account or become notify-only and the connection ends; lose the right to make changes and it becomes read-only.
  • Same limits as the API. Tool calls share the REST API's limit of 60 requests a minute per key or connected app, and run with your own permissions, so Claude sees exactly what you can see.
  • Upcertain is its own authorisation server. Tokens issued to Claude work only for the MCP server, not for the rest of the platform.

What it is not

Claude reads what Upcertain knows: monitors, checks, incidents and status pages. It does not see your logs or servers, so "why did it fail?" gets the recorded cause (a 503, a timeout, a missing keyword) rather than a root-cause analysis. Combine it with the tools that hold your logs if you want both in one conversation.

Prefer scripts?

The same data is available from the REST API with an API key, which owners and admins create. Both are described on the MCP server page.

Related