PagerDuty
by PagerDuty
Read incidents, on-call schedules and services — and with write scopes granted, create incidents that page whoever is on call. 72 tools, read-only default.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Connect PagerDuty via MCP
https://mcp.pagerduty.com/mcpWorks in any MCP-compatible client. In Agentman Studio it is one click — no config file to edit.
Use in Agentman
Connect PagerDuty once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.
Open in Agentman StudioPagerDuty Tools & Capabilities (64)
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Write access can page real people, with no in-protocol confirmation.
create_incidenttriggers the escalation policy and notifies on-call;add_respondersfires a named person's high-urgency notification rules. Neither tool's description mentions this. - The hosted server has no tool filtering. PagerDuty states
mcp-proxyfiltering is unavailable remotely, and theenable-write-toolsread-only default is a local flag. Scoped OAuth is the only remote control, and it fails calls rather than hiding tools. - We could not verify tool behaviour ourselves. The endpoint returns 401 to an anonymous handshake, so no safety annotations were readable. Read/write classification is PagerDuty's, from its documentation, not ours from the wire.
- PagerDuty's own docs disagree on the tool count. The tools overview says 72 tools across 17 domains; the tool-filtering page says 55 across 14; the write-tools page counts 28 write tools where the overview counts 22. We used the tools overview, because it enumerates every name and the enumeration itself totals 72 with 22 writes. The other two pages appear not to have been updated.
delete_teamhas undocumented downstream effects. Nothing states what happens to schedules, escalation policies or services that reference a deleted team.- Status page posts may be public and immediate. Whether a post is staged for approval depends on a per-page account setting the connector neither reads nor controls.
- Advanced Permissions is a hard prerequisite, which excludes Free-plan accounts and any Professional-or-above account that has not had the feature enabled by PagerDuty Support.
- No dynamic client registration. OAuth clients must be created in the PagerDuty account first, which is a deliberate administrative step rather than a self-service connect button.
- An account-level credential is not bounded by a person's role. The setup guide permits account-level credentials, and Advanced Permissions role limits do not constrain them the way they constrain a user token.
- Anthropic's snapshot lists no prompts, so there are no vendor-authored workflows to lean on; everything is driven from tool calls.
Frequently asked questions
Yes. PagerDuty's incident documentation states that manually opening an incident triggers it and notifies the on-call responder(s). An incident created by an agent escalates through the service's escalation policy exactly like one a human declares, reaching whoever is on call by push, phone, SMS or email according to their notification rules. There is no draft or staged state.
Yes. PagerDuty documents that responder notifications prompt users to accept or decline an invitation to join an incident, and that added responders receive notifications matching their high-urgency notification rules. So a single agent tool call can ring a named individual's phone regardless of the incident's own urgency, and that person must then explicitly accept or decline.
It depends on an account setting the connector does not control. Each external status page carries a Posts Approval option. Where no approval is needed, a post goes live to the public page immediately. Where approval is required, up to fifteen named users review the draft before it appears, and they receive an email notification that something is pending.
The self-hosted server is; the hosted one is governed differently. PagerDuty documents that its server runs in read-only mode by default, with 22 write tools behind an `enable-write-tools` flag. That flag is a local process argument. On the hosted endpoint the equivalent control is a scoped OAuth client, because PagerDuty states proxy-based tool filtering is unavailable remotely.
Whatever the connecting credential carries. PagerDuty's support documentation directs users to generate a user API token, so the agent acts as that person and inherits their base role and team roles. The setup guide adds that both user and account-level credentials are supported, and that some tools need user-level authentication to resolve filters like assigned incidents.
One that includes Advanced Permissions. PagerDuty's support page lists Advanced Permissions as a requirement, and that feature is available on Professional, Business, Digital Operations (legacy) and Enterprise for Incident Management plans. Accounts on those plans without it enabled must contact PagerDuty Support to request enablement. Free-plan accounts handle permissions through a different model entirely.
The directory snapshot is a subset, not a stale list. Every one of the 64 names Anthropic published on 2026-08-16 still appears in PagerDuty's current tools reference. The eight missing from the snapshot are all read-only: five analytics tools, two business-service tools and `list_priorities`. No listed name has been renamed or removed.
Yes, `delete_team` is one of the 22 documented write tools. PagerDuty's tools reference describes it only as deleting a team and states no consequence for schedules or escalation policies that reference it, so we could not establish that from the documentation. Treat it as unbounded until tested, and withhold `teams.write` from any agent that does not need it.
Sources
- PagerDuty MCP Server (support) — https://support.pagerduty.com/main/docs/pagerduty-mcp-server (last updated 16 June 2026; fetched 2026-08-19 as Markdown via the
.mdsuffix) · retrieved 2026-08-19 - MCP Tooling & Remote Server (developer portal) — https://developer.pagerduty.com/docs/mcp-tooling-remote-server (fetched 2026-08-19; the directory's
documentationURL, served as a client-rendered single-page app that returns an identical 539KB shell for every path, including a.mdsuffix) · retrieved 2026-08-19 - PagerDuty MCP Server Tools Overview — https://pagerduty.github.io/pagerduty-mcp-server/docs/tools/overview (fetched 2026-08-19; enumerates 72 tools, 50 read and 22 write, across 17 domains) · retrieved 2026-08-19
- PagerDuty MCP Server Incidents tools — https://pagerduty.github.io/pagerduty-mcp-server/docs/tools/incidents (fetched 2026-08-19) · retrieved 2026-08-19
- PagerDuty MCP Server Write Tools — https://pagerduty.github.io/pagerduty-mcp-server/docs/configuration/write-tools (fetched 2026-08-19; read-only default and the
enable-write-toolsflag) · retrieved 2026-08-19 - PagerDuty MCP Server Tool Filtering — https://pagerduty.github.io/pagerduty-mcp-server/docs/configuration/tool-filtering (fetched 2026-08-19; states filtering is unavailable for the remote server, and gives a conflicting count of 55 tools across 14 domains) · retrieved 2026-08-19
- PagerDuty Remote MCP Server Setup — https://pagerduty.github.io/pagerduty-mcp-server/docs/remote-server/setup (fetched 2026-08-19; EU endpoint, credential options, no dynamic client registration, scoped-OAuth filtering) · retrieved 2026-08-19
- Incidents — https://support.pagerduty.com/main/docs/incidents (fetched 2026-08-19 as Markdown; establishes that manually opening an incident triggers it and notifies the on-call responders) · retrieved 2026-08-19
- Add Responders — https://support.pagerduty.com/main/docs/add-responders (fetched 2026-08-19; responder requests use high-urgency notification rules and require accept or decline) · retrieved 2026-08-19
- External Status Page — https://support.pagerduty.com/main/docs/external-status-page (fetched 2026-08-19 as Markdown; Posts Approval setting, subscriber notifications) · retrieved 2026-08-19
- Advanced Permissions — https://support.pagerduty.com/main/docs/advanced-permissions (last updated 14 July 2026; fetched 2026-08-19 as Markdown; plan availability and base roles) · retrieved 2026-08-19
- Live RFC 9728 protected-resource descriptor — https://mcp.pagerduty.com/.well-known/oauth-protected-resource/mcp (fetched 2026-08-19; the root form 404s and the path-insert form returns the 401 challenge) · retrieved 2026-08-19
- Live RFC 8414 authorization-server metadata — https://mcp.pagerduty.com/.well-known/oauth-authorization-server (fetched 2026-08-19; scope list byte-identical to the resource descriptor's) · retrieved 2026-08-19
- Live anonymous probe — 2026-08-19: an
initializePOST returns HTTP 401 withwww-authenticate: Bearer error="invalid_token", … resource_metadata=…; GET returns 401 · retrieved 2026-08-19 support.pagerduty.com/main/llms.txt— a machine-readable documentation index PagerDuty points at from every support page (fetched 2026-08-19).developer.pagerduty.comserves norobots.txtand nollms.txt; both paths return the same SPA shell, and noContent-Signalheader was present on any PagerDuty host · retrieved 2026-08-19- Anthropic connector directory — https://claude.ai/directory/fc41b5e5-97bc-4c80-af18-e3a9442b655b (snapshot 2026-08-16) · retrieved 2026-08-16
Server Info
- Category
- Developer Tools
- Developer
- PagerDuty
- Tools
- 64
- Domain
- mcp.pagerduty.com
Using Claude Desktop or another MCP client? Setup docs — the connection URL above works anywhere.
Similar MCP Servers for Developer Tools
Related Resources
Ready to connect PagerDuty?
Connect PagerDuty once and your agents call these tools on their own — on a schedule, in a workflow, with nobody at the keyboard.