How DVT MCP waveform and log tools power AI-driven test failure root cause analysis

Overview

This video demonstrates how DVT MCP Log and Waveform tools power AI-driven root cause analysis for failing UVM tests. DVT AI Assistant's Root Cause Analysis profile orchestrates an AI agent that parses simulation logs, explores waveforms, and navigates source code to produce a structured, evidence-backed failure report. The same capability is also available as a skill for any AI agent connected to the DVT MCP Server, including Cursor, Claude Code, Codex, Copilot or Kiro.

Details

Introduction

In this video we will explore a particularly powerful AI application in which the DVT MCP Log and Waveform tools play a central role: performing Root Cause Analysis for a test failure.

To debug a failing UVM test, you typically navigate from the UVM error in the log to the corresponding source code, trace the issue across the testbench and design, and investigate signal evolution and causal relationships in the waveform.

AI Assistant Root Cause Analysis profile

The Root Cause Analysis profile in DVT AI Assistant drives this entire workflow by orchestrating an AI agent equipped with DVT MCP tools for log analysis, waveform exploration and code navigation. We simply ask AI Assistant to investigate the test failure in this project.

Notice how the agent first locates the failing test, then positively identifies the log and waveform files which is a mandatory checkpoint.

It then calls the dvt_get_simulation_log_messages tool, which parses the simulation log into structured UVM messages. Using various sorting and filtering criteria the model instantly retrieves the UVM errors, and then narrows down log messages to capture the events leading to the failure.

Then the agent searches for relevant signals in the waveform dump using hierarchical paths from the error context. Signal value changes are traced around the failure timestamp to reconstruct the timeline of events.

It also explores the source code referenced in the error messages using compiler-backed navigation, understanding both the verification environment structure and the design hierarchy.

The result is a structured report with a chain of events linking log messages to waveform evidence by timestamp, followed by a step-by-step explanation of the failure. Each claim includes clickable references so you can quickly correllate the provided information.

Root Cause Analysis Skill

The same root cause analysis capability is also available as a skill which can be used with any AI agent like Cursor, Claude Code or Codex connected to the DVT MCP Server.

Let's first insall the DVT MCP companion skills in $HOME/.agents.

Then we start the server in the project directory, start our agent and ask it to root cause the test failure.

The skill goes through the same stages as the DVT AI Assistant profile: parsing the simulation log, loading the waveform, searching signals and exploring source code, eventually producing a structured root cause report with evidence-backed conclusions.