Uptimy

AI Handover

Hand an incident to your own AI agent: Uptimy gives it the incident and how to investigate it, and your agent comes back with the likely cause and a plan.

Overview

Uptimy knows that something broke and where it was seen from: which regions failed, the error, the latency, and whether a service you depend on is down at the same time. Your AI agent (Claude Code, Cursor, Codex and the like) has what Uptimy doesn't: your terminal, your cluster, your logs and your code.

AI Handover joins the two. Every incident has an Investigate with AI button that gives you a prompt for your agent. The agent reads the incident from Uptimy, looks into your systems with the access you already gave it, and replies with the likely cause, what it ruled out, and a remediation plan for you to approve.

Uptimy doesn't run a model for this and never reads your logs. The investigation happens in your agent, with your access. It's on every plan.

Investigate an incident

  1. Open the incident on the Incidents page and click Investigate with AI.
  2. Click Copy prompt.
  3. Paste it into your agent: Claude Code, Cursor, Codex, or any other agent with access to your systems.

The button is on resolved incidents too. There the prompt asks for the root cause and what would stop it happening again, for the postmortem.

💡 Connect the Uptimy MCP first

The prompt works without it, from the details it carries. With the Uptimy MCP server connected, your agent also reads the incident's full timeline, the failed checks per region and the status of the services you follow, and can run a check to confirm a fix.

What the prompt contains

  • The incident's ID, severity, status and when it opened (and resolved), the monitor it's about, and how it failed: the error type and the regions that saw it.
  • The incident title and last error, quoted and marked as data, not instructions. Error messages and response bodies come from your systems and third parties, so a crafted one can't steer your agent.
  • The steps: read the incident, the monitor and its failed checks over MCP, rule out a provider outage, then investigate with the agent's own tools. The steps fit the incident: a website or API check, a cron heartbeat, a third-party service you follow, or an Uptimy Agent that went offline.
  • What to reply with: the likely cause with its evidence, what was ruled out, a remediation plan with exact commands (riskiest last) and how to confirm the fix.
  • To work read-only, ask you before running anything that changes something, and post its findings to the incident only as an internal update, never a public one, and only if you agree.

⚠️ Your agent acts with your access

The prompt asks for a read-only investigation and for your approval before any change, and most agents ask before running commands. It is still an instruction, not a permission. The key the Uptimy MCP creates when you connect has full access to the workspace, and your agent's access to your cluster and cloud is whatever you gave it. Review what it proposes before you approve it.

Automatic handover

A workflow can hand every new incident to your agent before anyone is paged. The prompt is available to every destination as {{event.handoverPrompt}}, the same text as the button. Three templates in the workflow builder use it:

  • Slack: AI handover posts the prompt when an incident opens, ready to paste into your agent.
  • GitHub issue for your coding agent opens an issue with the prompt as its body, labelled uptimy and ai-handover, and closes it when the incident resolves. Assign it to a coding agent that works on issues, or add the label it watches.
  • Webhook: start an agent run sends the prompt to your own runner, for example a CI job that runs an agent headless:
Webhook body
{
  "incidentUuid": "{{event.incidentUuid}}",
  "severity": "{{event.severity}}",
  "title": "{{event.title}}",
  "prompt": "{{event.handoverPrompt}}"
}

Values in a JSON webhook body are escaped as they're filled in, so the prompt's line breaks and quotes arrive as valid JSON.

Workflow tests are real runs: testing one of these templates posts the message, opens the issue or calls your webhook, with a sample incident.