Worked on the grafana/mcp-go repository to address a critical reliability issue in backend server shutdown. Focused on backend development and concurrency using Go, the work involved fixing a race condition in the SSEServer shutdown process. By carefully adjusting mutex unlock timing, the solution ensured the HTTP server could be accessed safely and ListenAndServe was invoked before releasing locks, enabling a more graceful and predictable shutdown. This change improved deployment reliability and reduced the risk of stuck connections during server lifecycle events. The update enhanced code maintainability and traceability, contributing to more stable on-call operations and smoother production deployments.