Oliver Kieselbach
066b6e6f04
Measured over 14 days, GET health/mcp answered in 4..148ms when warm and timed out at 30s in 50 of 83 calls — the median call took 27,972ms. A cold start is 20-30s of Container Apps activation (~98% KEDA plus image mount, ~500ms application), so no UI wait covers it and the generous timeout only bought a long wall in front of the same answer. The browser's own budget is the same 30s, so the backend's carefully worded warning never reached the user at all: they saw a raw DOMException. Backend: the budget drops to 3s, which sits in the widest observed latency gap (2347ms..3930ms) so jitter cannot flip the verdict. Our own cancellation now reports the new status "warming" at Information level (with minReplicas=0 a cold probe is the expected path, and logging it as a warning poisons the ops signal); HttpRequestException is split out of that catch and stays a real warning, still without the host in the message. The warm-hit threshold drops to 750ms — ~5x the warm maximum, deliberately generous because every measurement comes from a near-idle single-user system and warm latency will rise under concurrency — and moves into McpReachableMessage so a test can pin the boundary. The XML doc now describes the probe that exists rather than the wake-signal design the measurements disproved. "warming" is a new status value rather than an added DTO boolean because the backend deploys before the web: an unknown value degrades to a grey card with a clean banner, whereas a boolean would leave old web showing yellow and "Warnings detected" on every cold start. HealthCheck.Status now documents the five-value vocabulary, which rides into both generated TypeScript mirrors as JSDoc. Web: resolveMcpCardState owns display, banner rating and the poll decision as a pure function (the suite runs on node, so testable logic cannot live in JSX). A warming server re-probes every 5s for up to 8 attempts and stays out of the overall banner exactly as "unknown" already does — scaling from zero is expected behaviour, not a degraded platform — and only becomes a real warning once the re-polls are spent. The card keeps showing "warming" while the next probe is in flight so it cannot flicker, carries the blue in-progress look instead of yellow, and was hoisted out of the healthResult guard: a failing detailed check would otherwise hide a card that is still polling. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Autopilot Monitor
Autopilot Monitor is an open-source, real-time monitoring and troubleshooting platform for Windows Autopilot and Microsoft Intune enrollment. It records enrollment events, application installations, policies, device state, and diagnostics while provisioning is still running, instead of collecting logs by hand after a failure. Administrators get live enrollment tracking, Enrollment Status Page (ESP) visibility, detailed session timelines, automated issue detection, on-demand diagnostics, fleet health analytics, and AI-assisted troubleshooting through a hosted Model Context Protocol (MCP) server.
The platform supports classic Windows Autopilot and modern Intune provisioning scenarios, including Windows Autopilot Device Preparation.
Availability
Autopilot Monitor is publicly available — the Community plan is free. Sign in with your work account at autopilotmonitor.com; new organizations complete a short activation step after first sign-in.
Key Features
- Real-time Windows Autopilot monitoring — follow device provisioning while enrollment is still running
- Detailed enrollment timelines — correlate Autopilot, ESP, Intune Management Extension, apps, policies, device state, and system events
- Enrollment Status Page visibility — see which phase, application, or policy is delaying provisioning
- Automated issue detection — define custom analysis rules that identify known failure patterns
- Targeted evidence collection — gather registry values, event log entries, files, log-parser matches, and allow-listed commands when specific conditions are detected
- On-demand diagnostics — request diagnostics packages from enrolled devices remotely
- Fleet health analytics — success rates, deployment duration, device models, applications, errors, and recurring issues
- Session reporting — inspect and share detailed enrollment information for individual devices
- Windows Autopilot Device Preparation support — monitor Device Preparation deployments alongside classic Autopilot scenarios
- Centralized agent configuration — control diagnostic behavior, automatic reboots, timezone adjustment, and other monitoring settings from the web dashboard
- AI-assisted troubleshooting — query enrollment data through the built-in Model Context Protocol server
How it works
Autopilot Monitor consists of several components working together throughout Windows enrollment:
- Bootstrap Script — PowerShell script deployed through Microsoft Intune that starts monitoring early in the provisioning process
- Monitoring Agent — lightweight .NET application that collects telemetry and evidence during enrollment
- Backend API — Azure Functions-based ingestion and processing pipeline
- Web Dashboard — Next.js application for real-time monitoring, troubleshooting, and fleet analytics
- MCP Server — Model Context Protocol server that lets AI assistants query Autopilot Monitor data in natural language
Windows Autopilot and Intune Troubleshooting
Autopilot Monitor helps investigate common Windows provisioning and enrollment problems such as:
- Windows Autopilot deployments that fail or appear stuck
- Enrollment Status Page (ESP) delays and unusually long deployment phases
- Win32 application installation failures and slow application downloads
- Intune Management Extension problems
- Microsoft Entra join, Hybrid Join, and device registration issues
- policies that block or delay provisioning
- inconsistent behavior across hardware models and recurring failures affecting multiple devices
Because telemetry is collected during enrollment, many problems can be investigated without reproducing the issue or manually collecting logs from the affected device.
AI Integration (MCP)
Autopilot Monitor ships a hosted Model Context Protocol server. Connect Claude Desktop, VS Code, or any MCP client that speaks Streamable HTTP with OAuth, and ask questions like:
- "Show me all failed enrollments from the last 24 hours and group them by failure reason."
- "Why did session X fail, and which app install caused the delay?"
- "Which devices had unusually long ESP application phases this week?"
- "Which device models have the highest Autopilot failure rate?"
- "Which devices in my fleet are affected by CVE-2024-30078?"
Server URL: https://mcp.autopilotmonitor.com/mcp — sign-in runs through your existing work account, access is scoped to your tenant exactly like in the portal, and the MCP server stores no credentials. Setup guide: AI Integration (MCP).
Documentation
Full admin documentation is available at docs.autopilotmonitor.com — deployment, configuration, monitoring, diagnostics, troubleshooting, analysis and gather rules, integrations, and AI-assisted troubleshooting.
Open Source
Autopilot Monitor is developed openly on GitHub. The source is published so that customers and security reviewers can see what runs on their devices and with their data: the repository contains the device agent, shared library, backend, web dashboard, and MCP server.
Feedback
Bug reports and feature requests go through Issues, questions and ideas through Discussions. Security issues are reported privately, see SECURITY.md. For a problem in your own tenant, the Report Session button in the portal is the fastest path.
License
This project uses a split licensing model. The root LICENSE is AGPL-3.0, which covers the server-side components; the device-side components carry their own MIT license file:
- MIT License — Agent (
src/Agent/) and Shared library (src/Shared/) — unrestricted use on end-user devices, no copyleft obligations - AGPL-3.0 — Backend (
src/Backend/), Web Dashboard (src/Web/), and MCP Server (src/McpServer/) — modifications to server-side components remain open source, especially when deployed as a network service
The Shared library is MIT because it is a dependency of the MIT-licensed Agent. The license file inside each component directory is authoritative for that component.