---
title: "MCP: Root Cause Analysis"
url: "https://atlan.com/demos/mcp-root-cause-analysis/"
excerpt: "Trace a broken dashboard to its root cause in one pass with the Atlan MCP—no more manual tool-hopping across layers."
description: "A dashboard breaks. Instead of hopping between your BI tool, pipelines, transformation layer, and source tables hoping to land on the failure, point the Atlan MCP at it and let the agent do the tracing. Prompt it with the alert. It confirms the right asset, walks lineage upstream through every layer, and surfaces each failure signal tied to the exact asset where it lives—then hands you a structured root cause report you can act on immediately. No more context switching. No more 20 minutes of checking jobs, querying logs, and cross-referencing systems. One pass, one clear answer, with enough context to explain it to business users. The same pattern extends to tables, data products, and pipeline assets. And once you trust the output, automate it—so the RCA is already waiting when the first alert hits."
format: "Video"
duration: "PT3M10S"
video: "https://videos.ctfassets.net/nwa1c00rtgxb/7iYosd2u6vC9epZrNOTtnw/808563fb52a742211f54631c6fcb34a8/Root_Cause_Analysis.mp4"
thumbnail: "https://images.ctfassets.net/nwa1c00rtgxb/1bSI2pXe9hj3iIyKp3ZSTi/eedd59a3b365936486a472060b6ab821/mcp-root-cause-analysis.webp"
content_purpose: ["Product Overview"]
target_persona: ["General", "Data Engineer"]
journey_stage: ["S1 - Prospecting", "S2 - Discovery", "S3 - Solution Design", "C1 - Onboarding", "C2 - First Value", "C3 - Adoption"]
use_case_context: ["Training"]
product: ["Context For AI - MCP"]
published: "2026-06-03"
updated: "2026-06-08"
content_type: "video transcript"
transcript_source: "sheet"
---

# MCP: Root Cause Analysis

Transcript of the video at https://atlan.com/demos/mcp-root-cause-analysis/

─ When a dashboard breaks, the instinct is to start checking everything manually. The BI tool, the pipeline, the transformation layer, the source tables one tool at a time, hoping you land on the failure before someone escalates. In this walkthrough, we'll use the Atlan MCP to trace upstream lineage from a broken dashboard, surface every failure signal along the way, and generate a structured root cause analysis report so the team has a clear answer fast, not after an hour of tab switching. ─In this case, we start in our LLM and prompt it with a simple alert that we got. ─ That something is wrong with a specific dashboard and we need to investigate what may be causing it.

Once submitted, the agent picks that up as its instructions and it's ready to begin the investigation. The agent first brings back the asset that it has found using the Atlan MCP, and presents it to us with all the metadata associated with it. This allows us to ensure that we are looking at the correct asset. Because the Atlan MCP surfaces all the metadata to us directly in our conversation. It gives us the context we need to ensure we are investigating the right asset. We confirmed that this is correct and the agent starts walking lineage.

Following the lineage graph upstream, the agent walks through every layer from the bi layer back to the transformation layer and into the source tables. It's covering the ground. That would take a data engineer, switching applications and navigating UI just to replicate. ─────Each failure signal the agent finds is tied to a specific asset in the lineage path. That's important context. You're not just getting a list of errors, you're getting a map of which upstream layer each failure lives in. That's what makes the next step possible. With the lineage reversal complete and the failure signals collected, the agent assembles the report.

It names the root cause, the specific asset or layer where the failure originated and outline what needs to happen to resolve it. The report lands directly in our conversation for us to review before anything else is done. ─What this workflow replaces is the manual tool hop the 20 minutes of checking jobs, querying logs, and cross-referencing dashboards across systems to piece everything that went wrong together. ─ The agent traverses the lineage and surfaces every failure signal in one pass. The engineer's judgment stays in the loop at the review step where it belongs.

The result is a root cause report The team can act on immediately with enough context to explain it clearly to the business users. ─ If you wanna extend this further, the same pattern works beyond dashboards, tables, data products, and pipeline assets all support the same lineage reversal approach through the Atlan MCP. And once your team has validated the workflow on a few real incidents, and you trust the agents' output, you can configure it to run the RCA automatically on failure detection. So the report is already waiting when the engineer opens their first alert. ──