PagerDuty MCP Server

The PagerDuty MCP Server connects an AI assistant to your PagerDuty account. Instead of switching between dashboard, chat, and terminal during an incident, you ask directly from your development environment: "Which incidents are open right now and who is on call?", "Show me the details of this incident", or "Which team owns the checkout service?". The assistant calls the matching tools and reads or changes data in your real account. PagerDuty itself operates and maintains the server.

What is the PagerDuty MCP Server?

The server implements the Model Context Protocol (MCP) and, according to the provider, runs as a managed service at PagerDuty. You install nothing locally; you enter the endpoint https://mcp.pagerduty.com/mcp in your MCP client, and accounts in the EU region use https://mcp.eu.pagerduty.com/mcp. The earlier local variant with uvx pagerduty-mcp is deprecated according to the README, and the repository is archived and explicitly points to the hosted server. If you come across an older tutorial, use PagerDuty's current documentation instead.

Features and tools

The tools cover the core of the platform. According to the documentation, you can search and manage incidents, view services and teams, query on-call schedules and escalation policies, work with alert grouping and event orchestrations, and look at status pages and users. The tools are largely built as pairs for listing and managing, so reading and changing can be told apart. This is especially useful during troubleshooting: the assistant can fetch the incident, the affected service, the responsible team, and the current on-call person one after another and turn that into a compact situation summary.

Setup and sign-in

The prerequisite, according to the provider, is a PagerDuty account with Advanced Permissions. Sign-in works either with a personal API key in the header Authorization: Token token=... or through OAuth with a dedicated OAuth client. Dynamic Client Registration is not supported, so you have to set up the client in PagerDuty beforehand. In Claude Code, a command such as claude mcp add-json pagerduty with URL and header is enough; for Cursor and VS Code there are configuration examples in the documentation. An OAuth client with a restricted scope additionally limits which tools can be used.

Costs and licence

Access to the hosted MCP server requires a PagerDuty account whose cost depends on the plan you choose. PagerDuty publishes current terms on its official pricing page. The source code of the earlier local variant is licensed under Apache 2.0, while the hosted service itself is governed by PagerDuty's terms of use.

Who is the PagerDuty MCP Server for?

It suits on-call and platform teams that already use PagerDuty and want to bring incident knowledge into their workflow faster, for example developers who want to know while debugging which incident belongs to an error. An assistant also saves clicking when tidying up services, teams, and escalation rules. If you do not use PagerDuty, you gain nothing. For very critical escalation changes, the web interface with its confirmation logic remains the clearer choice.

Security and limits

The server acts with the permissions of the API key or OAuth client used. An assistant can therefore change incidents and adjust configuration. If possible, use an OAuth client with a narrow scope and confirm write tools one by one in the client. Also keep in mind that incident texts and service names enter the context of the AI model. PagerDuty's documentation recommends validating authorisation regularly.

Frequently asked questions

Is the local uvx variant still current? No, according to the README it is deprecated and the repository is archived. Use the hosted server.

Which address applies to EU accounts? Accounts in the EU region use https://mcp.eu.pagerduty.com/mcp.

Do I strictly need OAuth? No, a personal API key is enough, but OAuth allows finer limiting of the tools.

PagerDuty account with Advanced Permissions, a personal API key or an OAuth client, and an MCP client with support for HTTP servers (Claude Code, Cursor, VS Code).

The hosted PagerDuty MCP Server lets AI clients such as Claude Code or Cursor query incidents, services, and on-call schedules. This guide shows the fastest path to a first working connection.

Step 1: Check the prerequisites

You need a PagerDuty account with Advanced Permissions and an MCP client that supports HTTP servers. Clarify beforehand whether your account is in the US or the EU region, because the endpoint differs.

Step 2: Choose the endpoint

US accounts use https://mcp.pagerduty.com/mcp, EU accounts use https://mcp.eu.pagerduty.com/mcp. The earlier local variant with uvx pagerduty-mcp is deprecated and its repository is archived; do not set it up anew.

uvx pagerduty-mcp

Step 3: Sign in with an API key

Create a personal API key in PagerDuty and store it as the environment variable PAGERDUTY_API_KEY. In Claude Code, add the server with claude mcp add-json pagerduty, using type http, the URL, and the header Authorization: Token token=${PAGERDUTY_API_KEY}.

claude mcp add-json pagerduty

Alternative with OAuth

For finer permissions, create an OAuth client with a limited scope in PagerDuty and store its client ID and secret in the client configuration. Dynamic Client Registration is not supported, so the client must exist beforehand.

Step 4: Test the connection

Start with a read-only request, such as "Which incidents are currently open?" or "Who is on call today for the platform team?". Only once that works is it worth moving on to write actions.

Set limits deliberately

The server acts with the full permissions of the key. Confirm write tools one by one and, if possible, use a narrowly scoped OAuth client.

Personal API key in the header Authorization: Token token=... or OAuth with a pre-created client (Dynamic Client Registration is not supported).

Permissions follow the API key or OAuth client. The tools can read and modify incidents, services, teams, schedules, and escalation policies.

The server is hosted by PagerDuty; requests go through its endpoint. Incident details, service names, and on-call data enter the AI client context and therefore reach the respective model provider.

Write actions on incidents and escalation rules, an API key stored in plain text, overly broad OAuth permissions, and sensitive operational data in incident texts.