HYBE Builds Automated Incident Investigation System Using AWS DevOps Agent and Custom MCP Server

When an incident occurs, operators have to switch context by moving between scattered tools such as Datadog, GitLab, AWS, and Jira. In particular, since system-specific naming differs from service to service, identifying the cause alone took an average of 30 minutes to 1 hour, creating inefficiency.

To solve this, HYBE's Infrastructure Operations Team introduced a hybrid strategy combining AWS DevOps Agent with a Custom MCP (Model Context Protocol) server. By adding a Custom MCP server that connects in-house tools to the autonomous investigation and topology analysis capabilities of AWS DevOps Agent, they automated the entire workflow.

The system that was built handles the entire process from error detection to autonomous investigation, automatic Jira issue creation, and after the issue is resolved, MR creation and automatic review. The four Custom MCP servers were developed in-house using Python FastAPI and deployed on EKS, and the system was successfully put into production in just 6 weeks.