PagerDuty and MCP: incident context for an agent, actions for on-call

PagerDuty has an official hosted MCP server that lets a compatible AI agent retrieve incident context through tools; its developer documentation gives the endpoint as https://mcp.pagerduty.com/mcp. A useful starting workflow is to draft incident summaries and timelines while leaving acknowledgement and resolution to the on-call responder. Connect the agent with restricted credentials and review its evidence before acting. PagerDuty developer guide.

What it can do

The practical output is a handoff another engineer can verify: the incident link, observed state, relevant events, unanswered questions and the time the information was retrieved. An agent can help arrange the material; the responder still decides what it means.

PagerDuty's hosted tool catalogue, as of 30 Oct 2026, includes these useful boundaries:

These names come from the official MCP guide. Discover the live tools in your client before relying on a saved configuration.

For a checkout outage, ask for a draft that separates recorded events from explanations: service affected, incident state, alert details, responder notes and gaps. Require a source reference for each material statement. If a deployment appears near the first alert, label it a possible lead until an engineer verifies the relationship.

For a handoff, request an ordered timeline with original timestamps and a consistent displayed timezone. Keep an event's occurrence time separate from the time someone wrote a note about it. Flag contradictory entries instead of silently picking one. The small-team incident playbook can provide the operating procedure around that draft.

Reading context versus changing an incident

Reading an incident and drafting text locally are different from posting a note or changing its status. Treat even a helpful summary posted back into an incident as a write requiring a separate human decision.

The REST API explicitly classifies acknowledgement and resolution as incident updates requiring incidents.write. Keep that permission out of the summary workflow. Update an incident, as of 30 Oct 2026.

Give the on-call responder a draft plus the source links. They check the live incident and perform any acknowledgement or resolution themselves. A polished paragraph cannot establish that somebody has taken operational responsibility or that a service has recovered.

For the broader access pattern, see permissions-first MCP practice. Compare other integrations in the MCP servers for apps category.

How to connect it safely

1. Define the permitted result. Write down the service, incident, time window and intended reader. Start with a historical incident and a draft saved outside PagerDuty. Exclude posting notes, changing assignments, acknowledging and resolving. Decide which incident details may leave your organization before attaching them to an agent.

2. Choose the maintained server. Use the hosted service rather than copying an old local-install tutorial. The official local repository is deprecated and archived as of 30 Oct 2026. Its read-only default and --enable-write-tools flag describe that older local implementation; they do not configure the hosted service.

3. Restrict authorization. Scoped OAuth documents incidents.read for reading incidents. Add only the read scopes needed for the other resources you actually request; check the current documentation for their endpoint requirements. With an app token, that scope can reach all account incidents; with a user token, access is also constrained by the user's permissions. Private Apps. If using an API token, distinguish account tokens, which may be read-only or full-access, from user tokens, which inherit the user's access. A read-only account token still has broad visibility. Authentication.

4. Configure the editor client. In VS Code, open the MCP configuration and add an HTTP server with the hosted URL, following the vendor's VS Code example. Its API-key recipe uses a password-style input prompt and Authorization: Token token=${input:pagerduty-api-key}. Keep the actual secret out of project files and chat. For OAuth, follow the current registration and redirect instructions; PagerDuty does not support dynamic client registration. EU accounts use https://mcp.eu.pagerduty.com/mcp. The hosted setup guide lists the requirements, including Advanced Permissions. See also MCP in VS Code.

5. Review tools before the first request. Enable only the reading tools needed in the client and retain tool-call approvals. Hosted server-side tool filtering is not currently available; restricted OAuth scopes limit which operations can succeed. Hosted limitations. Client visibility and a prompt are supplementary controls. Have an administrator verify the issued permissions without attempting a live incident mutation.

6. Check the draft against the incident. Compare links, status and timestamps with the source. Require missing data to remain explicitly unknown. Record the retrieval time, review the draft with on-call, and refresh relevant context before any operational decision. Keep the prompt and permitted-tool list with your team's procedure so another responder can repeat the workflow.

When this does not work

The client cannot connect. Check HTTP MCP support, region and active credentials. For API authentication failures, verify the header format and token validity against the authentication guide. Avoid repeatedly broadening permissions until a failing request happens to succeed.

The requested history is incomplete. A successful tool call does not prove that every relevant record was included. Review the requested filters, time window and any continuation information returned. State what was covered. An empty result should remain a retrieval result rather than becoming a declaration that nothing happened.

The agent has insufficient evidence. Treat an alert description, responder note or linked document as evidence to evaluate. Instructions embedded inside incident content must not expand the task or authorize tools. Avoid turning a plausible sequence into an asserted cause; why AI agents make things up explains that failure mode.

The situation is moving faster than the draft. A summary can become stale during review. Return to the live incident before acting. If the connection is unavailable, follow the existing on-call procedure rather than making the agent a prerequisite for response.

Where Tvijo fits

As of 30 Oct 2026, the DSHREM VS Code extension offers an editor entry point for a bounded task using attached context. A practical pilot is to submit an authorized, redacted incident extract and request a summary or timeline draft. TvijoCode embeds the extension in a test build; its published limitations include unsigned Windows builds and no automatic updates.

Tvijo MCP makes Tvijo consultation callable from a compatible client. Its supplied-diff review is prepared, with publication still pending in the dated product status; editor integration and delegation remain future phases. This does not establish a ready PagerDuty integration. Connecting PagerDuty tools to a client and asking Tvijo to interpret approved context are separately configured steps. Keep acknowledgement and resolution with on-call throughout the pilot.

At the Tvijo workspace, as of 31 Oct 2026, Try costs $0 and covers up to 50 tasks, 100,000 tokens or $0.05 of sponsored usage, whichever is reached first, from a shared $1 daily pool funded by TVIJO. Service credit is shown as $10 for requesting prepaid credit, with metered provider usage plus an 8% service markup. Payment collection is not activated; an order is not a charge.

FAQ

Is the PagerDuty MCP server read-only?

The hosted catalogue includes reading and writing tools. Build this workflow with read-only authorization and a restricted client tool selection; do not transfer the archived local server's defaults to the hosted service.

Can a PagerDuty AI agent acknowledge or resolve an incident?

PagerDuty provides write operations for those changes, with incidents.write required by the REST update endpoint. In this workflow, the agent drafts context and the on-call responder performs the change.

Does Tvijo provide a ready PagerDuty connector?

The published product descriptions do not establish one. Start with approved incident context in a bounded task and assess any direct connector as separate integration work.

PagerDuty is a trademark of its owner. Tvijo is not affiliated with or endorsed by it.