Microsoft has recently faced scrutiny over the perceived sluggishness of Windows 11 and the pace of its enhancements. While some of this criticism may be unwarranted, the tech giant acknowledges that there are valid concerns, including ethical considerations that have impacted its modernization efforts.
Introducing ETW MCP for Enhanced Performance Analysis
In response to these challenges, Microsoft is forging ahead with the introduction of new tools aimed at modernizing Windows performance analysis. This week, the company unveiled a preview of the Event Tracing for Windows Model Context Protocol (ETW MCP), a server designed to streamline the analysis of Windows Event Tracing (ETW) logs through the integration of artificial intelligence.
The ETW MCP empowers developers by enabling tools like GitHub Copilot and other MCP-compatible assistants to examine Event Trace Log (ETL) files. These tools can answer queries related to the logs and assist in pinpointing performance issues, eliminating the need for developers to sift through extensive datasets manually.
For those unfamiliar with ETW, it serves as the built-in tracing infrastructure within Windows, collecting detailed performance and diagnostic data. Although ETW is a robust system, working with ETL trace files typically demands specialized knowledge and considerable effort, often requiring tools such as Windows Performance Analyzer (WPA). The introduction of ETW MCP aims to simplify this process by allowing developers to interact with traces using natural language.
It is important to note that raw trace events will not be directly accessible to AI models. Instead, the MCP server offers structured, read-only access to processed trace information. This functionality enables developers to query valuable insights, such as identifying which processes consumed the most CPU time, comparing multiple traces, examining thread activity, or uncovering critical execution paths. The heavy lifting of data processing is managed locally through Microsoft’s existing TraceProcessor engine, while the AI assistant focuses on interpreting the results.
Microsoft emphasizes the significance of this approach, as it ensures that responses are based on actual trace data rather than assumptions generated by models, which could lead to critical errors. Furthermore, because the MCP returns structured results, developers can request supporting evidence for any conclusions drawn by the assistant.
The potential applications for this new tool are numerous. Developers investigating performance regressions can swiftly identify bottlenecks, while hardware and driver engineers can analyze CPU usage or power-related challenges. Organizations can also integrate automated trace analysis into their testing pipelines or continuous integration environments. Notably, the server operates without a graphical interface, allowing for complete functionality from a terminal or scripted workflows.
Current features available in the ETW MCP preview include:
| Capability | What it does |
|---|---|
| Trace processing | Loads an ETL into a queryable session via TraceProcessor / DataLayer |
| Querying, filtering, aggregation | Bounded, structured results, no raw event firehose at the LLM |
| Multi-trace comparison | Baseline vs. trial deltas in a single conversation |
| Symbol-aware analysis | module!function, WPP, and managed symbols when symbol paths are configured |
| Critical Path Analysis (CPA) | The same CPA signatures we ship in the external XPerf NuGet |
| Regions of Interest | Score and compare named time windows across many traces |
Additionally, Microsoft has indicated that the ETW MCP will complement the forthcoming WPA MCP integration. While the ETW MCP is tailored for headless environments and automation, the WPA MCP will offer similar AI-assisted analysis directly within WPA for users who prefer a graphical interface.
As with other AI-driven developer tools, Microsoft advises that responses generated by large language models should be viewed as guidance rather than definitive conclusions. Developers are encouraged to validate findings against the underlying trace data and utilize follow-up queries to confirm any analysis provided by the assistant.
The ETW MCP preview is currently accessible as the Microsoft.Windows.EventTracing.MCP NuGet package. Microsoft is actively seeking feedback from developers regarding prompt quality, supported scenarios, performance on large traces, and additional analysis capabilities. The company has also confirmed plans for further enhancements and improved integration with the entire Windows Performance Toolkit in future releases.