Give Your JVM an AI Assistant: JFR, JDK Mission Control and MCP for Production Performance

You'll walk away able to: stop guessing why your Java app is slow, and let AI investigate with real JVM evidence.

Register NowPresented byRavi Soni

• ✓Capture and read JFR recordings in JDK Mission Control to find CPU, allocation, GC and locking problems
• ✓Expose JVM diagnostics as MCP tools an AI assistant can query
• ✓Walk from symptom to root cause to a fix, then prove it with a second recording
• ✓Keep engineers in charge: AI correlates the evidence, you validate the diagnosis

Your Java application is slow. What do you do next?

Production JVM performance investigations often begin with the same symptoms: rising latency, high CPU, increased allocation, GC activity, blocked threads, or unexpected throughput degradation. The challenge is not collecting telemetry—it is connecting the evidence quickly enough to identify the real bottleneck.

In this session, we explore a practical approach to giving your JVM an AI assistant by combining JDK Flight Recorder (JFR), JDK Mission Control (JMC), AI, and Model Context Protocol (MCP).

JFR provides rich runtime evidence from the JVM, while JMC provides powerful capabilities for analyzing CPU usage, memory allocation, garbage collection, threads, locks, and application behavior. MCP provides a way to expose these diagnostic capabilities as tools that an AI assistant can interact with.

Rather than asking an LLM to guess what is wrong, we will give the AI access to real JVM evidence and let it investigate a production-style incident. Through a hands-on Spring Boot scenario, we will move from an initial performance symptom to JFR capture, JMC analysis, AI-assisted investigation, root-cause hypothesis, remediation, and finally a second JFR recording to validate the improvement.

We will investigate common JVM problems including CPU hotspots, excessive object allocation, GC pressure, thread contention, locking, and latency. We will also demonstrate how MCP can turn JVM diagnostics into actionable tools for an AI assistant—allowing developers to ask questions such as “Why is this service slow?”, “What is consuming the CPU?”, “Are threads blocked?”, and “Did our performance fix actually work?”

The goal is not to replace JVM engineers with AI. It is to demonstrate a new evidence-driven performance engineering workflow, where JFR provides the facts, AI helps navigate and correlate those facts, and engineers remain responsible for validating the diagnosis and deciding what changes should reach production.

The key idea is simple: don't ask AI to guess why your Java application is slow. Give it the JVM evidence.

Ravi Soni

The CodeFather · Java & Cloud-Native Architect

Ravi Soni, known as The CodeFather, is AWS Certified and a Java and cloud-native architect with 15+ years of professional experience in the design, development, maintenance, and migration of Java, JavaEE, Spring Boot, Kafka, and Kubernetes-based distributed and microservice applications.