AI-Powered Incident Response System for Notion
The author built an agent that automates incident response in Notion by assembling a war room page with all the necessary information, and generating an AI-written post-mortem when the incident is resolved.
Why it matters
This agent demonstrates how AI and automation can streamline critical incident response processes, reducing response time and improving reliability.
Key Points
- 1Incident Runbook agent detects new high-severity incidents in Notion, then assembles a war room page with incident details, runbook, on-call contacts, and service dependencies
- 2Uses Notion's Model Context Protocol (MCP) to read from multiple databases and write the war room page without any direct API calls
- 3Handles relation resolution between databases and writes Markdown content back to Notion
- 4Includes a 'watch' mode that continuously monitors for new incidents
Details
The Incident Runbook agent was built for the Notion MCP Hackathon. It addresses the problem of manual, scattered incident response by turning Notion into a living incident management system. When a new high-severity incident is detected, the agent cross-references databases for affected services, runbooks, and on-call contacts, then assembles a comprehensive war room page with all the necessary information. This is done entirely through Notion's MCP without any direct API calls. The agent also generates an AI-written post-mortem when the incident is resolved, calculating MTTR and summarizing the incident details. Key technical challenges included normalizing relation IDs across databases and efficiently writing Markdown content back to Notion.
No comments yet
Be the first to comment