Dev News Daily ENDE

Quarkus AI Root Cause Analysis feeds Tempo, Loki and Prometheus data to an LLM over MCP

The Quarkus project has introduced AI Root Cause Analysis, an open-source Quarkus application that connects to an existing Grafana LGTM stack through the Model Context Protocol and asks a language model to explain what went wrong. Ales Justin described it on the Quarkus blog on 10 November.

The workflow follows the steps an engineer would take by hand. The tool fetches recent traces from Tempo and picks out those with errors, high latency or unusual patterns, pulls the logs for each trace from Loki and ties them to individual spans, and queries Prometheus for metrics in the same time window. The model then receives the combined picture and returns a structured report with root causes, severity, cascading failures and suggested next steps. Optionally it can read the source code of monitored applications through Quarkus Dev MCP and generate Grafana dashboards for the problems it found, written into each application's workspace.

The post calls the data pipeline the key part. Raw telemetry from MCP quickly fills a model's context window, so a client decorator filters each tool's output before it reaches the model: SDK metadata and localhost attributes come out of traces, duplicated fields out of logs, and Netty internals, constant values and histogram buckets out of metrics. For metrics the post reports a reduction from 586 entries to about 35.

In the example given, a gateway shows three-second responses; the tool finds connection-pool exhaustion warnings on the backend and JVM heap at 92 percent with frequent garbage collection, and reports the pool as the cause and the gateway latency as a symptom. The application can be run as an uber-jar from Maven Central against an existing stack, or in dev mode, where Dev Services start an LGTM stack automatically. Providers are selected through Maven profiles: OpenAI by default, plus Grok, Gemini and watsonx.

Quarkus AI Root Cause Analysis feeds Tempo, Loki and Prometheus data to an LLM over MCP
Quarkus AI Root Cause Analysis feeds Tempo, Loki and Prometheus data to an LLM over MCP — Dev News Daily

Why it matters

The useful part is not the model but the filtering in front of it. Whether a summary is right still depends on the telemetry it was given, so the report is a lead for an engineer to verify, not a verdict.

Written by Victoria Shinder.