[ONBOARD] Disk Performance Troubleshooting

### Service / Tool Name

Disk Performance Troubleshooting Skill

### Contacts

Harsha Koonaparaju (hkoonaparaju), Rakkimuthukumar Nallore Ponnusamy (Rakki.Muthukumar), Kenaz Kwa (kenazkwa), Lakshmi narasimha Viswanadha (lakvis), Christine Sun (christinesun)

### Intended Agent Scenarios

Problem: When disks attached to a VM perform poorly (IO throttling or increased latency), customer workloads take a hit. Today, one of our largest enterprise customers files a support ticket with the team for RCA every time this happens, and they've asked whether we can expose this information through an MCP server so their agents can self-serve and avoid tickets entirely. The impact is significant: disk performance cases account for more than 70% of all Disks-related support cases and have the highest average support engineer time of any disk scenario (290–320 minutes per case).

What this skill/tool does: For this we want to build a skill + tool in the MCP server. The tool collects the data for disk metrics and passes it to the skill. The skill decides the next steps based on the data the tool provides. At this point the decision will be: is it a disk problem or not, and if it's a disk problem we will provide next steps for the customer to take. Diagnostic logic is authored and reviewed by the Azure Disks product group.

Existing Azure MCP services don't include this: Azure Compute covers managed disk CRUD only, Azure Monitor exposes raw metrics without disk-specific analysis, and Azure App Lens provides generic diagnostic detector access but does not include a disk performance detector.

### Timeline

June 2026.