The PagerDuty MCP server is the official bridge between Claude and your incident-response stack. It exposes the PagerDuty API so an on-call engineer can triage from a chat window instead of clicking through the web console. If you run SRE or ops work and want an AI copilot during an incident, this is the direct line in.
What it actually does
The server surfaces the core PagerDuty objects: incidents, services, on-call schedules, and escalation policies. You can ask about what is happening right now, who is responsible, and how a service is configured, then take action without leaving the conversation.
Practical patterns:
- ‘List every triggered incident on the payments service and tell me who is on call.’
- ‘Acknowledge incident P123456 and summarize the recent notes.’
- ‘Who is the next escalation contact if the primary does not respond in ten minutes?‘
Why use it
During an incident, context switching is expensive. Pulling schedules, incident history, and service ownership into one place lets Claude assemble the picture faster than a human clicking between tabs. It is most useful for the first few minutes of triage, when you are still working out scope and ownership.
Gotchas
Auth is a PagerDuty API token, so scope it carefully. A token that can resolve or reassign incidents can do real damage if misused, and actions taken through the server are real PagerDuty writes, not a sandbox. Treat destructive operations (resolving, reassigning, escalating) with the same caution you would in the console. The server is only as current as the API, so a very fresh incident may lag by a moment.
