AI
The Cost of Context Switching
An incident alert hits. Slack messages fly. PagerDuty notifications pile up. The engineer opens tabs for GitHub, metrics, logs, and runbooks. Each switch loses train of thought. Focus recovery takes 30 minutes across five tools. Mistakes become more likely. Productivity drops. Rocío Bayon framed the problem: dragging developers out of their platform costs time and increases error rates. The fix is to bring all context to the IDE.
Four Agents Under One MCP Server
PagerDuty Advance ships four agents. The describe agent transcribes conference bridges into incident context. The chief agent manages schedules and overrides. The insights agent retrieves metrics. The SR agent pulls logs from Elasticsearch, CloudWatch, and other vendors, plus runbooks from Confluence and ServiceNow. All actions are recorded in the incident audit trail. The MCP server exposes these agents as tools. The IDE discovers them on initialization and presents them as MCP apps.
Demo: Triage and Fix Without Leaving VS Code
Sebastian Villanelo showed a developer paged for an API timeout error. In VS Code Chat, he listed incidents via the PagerDuty MCP server. A new MCP app, the Incident Command Center, displayed urgency, priority, alerts, payload, and timeline. Two buttons appeared: “Triage locally” and “Call SR agent.” The SR agent retrieved logs, checked a similar incident from the past 30 days, and returned a summary with severity, impact, and suggested next steps. Sebastian then applied a fix (restored a timeout to 60 seconds), deployed it, added a note, and resolved the incident.
Beyond VS Code: Cloud Desktop and Multi-Agent Future
The same Incident Command Center runs on cloud desktops (e.g., Goose). A demo showed triage with log retrieval from Elasticsearch directly in the cloud desktop view. PagerDuty also built a service dependency graph showing impacted services during an incident. An on-call compensation report tracks responder hours across business hours and holidays, configurable per country. Rocío Bayon outlined the next step: multi-agent workflows where the SR agent coordinates with a security agent or deployment agent, with the human supervising.
Q&A
How does the deployment agent know it can fix something? The agent-to-agent ecosystem is under development; the SR agent currently works with integrations for logs and may later perform actions. ▶ Watch (20:02)
What would you change about the MCP app spec? The biggest headache was the view layout — VS Code allows resizing, but cloud desktop uses a fixed canvas. A standard canvas for MCP apps would help. ▶ Watch (21:33)
Notable Quotes
I don’t want to leave the IDE I want to do the whole triage the whole UI that you have in the ID. Sebastian Villanelo · ▶ Watch (2:50)
the SR agent is going to pull information from some log vendors like elastic search, cloudatch and other specific integrations Sebastian Villanelo · ▶ Watch (4:09)
we have a rich UI in the ID that’s putting everything together in one place for us Rocío Bayon · ▶ Watch (16:24)
we want the AI to handle you know the chaos the toil and then have the human just make the critical decisions Rocío Bayon · ▶ Watch (18:25)
Key Takeaways
- MCP apps bring PagerDuty’s incident command center into VS Code, eliminating tool switching.
- The SR agent automates log retrieval, runbook lookup, and triage suggestions.
- Experimental MCP apps extend to cloud desktop, service graphs, and on-call compensation reports.
About the Speaker(s)
Sebastian Villanelo develops custom reports that help each customer identify and monitor the metrics most relevant to their operations. He gathers technical and functional requirements, working closely with the product team to translate customer needs into concrete improvements.
Rocío Bayon is a Product Manager on the Forward Deployed Engineering (FDE) team at PagerDuty. With a background in Mechanical Engineering and Business Analytics, she lives at the intersection of technology, data, and real-world customer implementations.