EdgeSentinel - an mcp which troubleshoots linux systems.

About this project

The Problem: The Edge is a Black Box Managing remote edge devices—IoT gateways and digital kiosks—is a logistical nightmare. When nodes drop offline, engineers must manually SSH in and grep through cryptic kernel logs to guess the failure. This archaic process is unscalable, and feeding raw OS errors to standard AI agents often causes hallucinations, rendering them useless for low-level infrastructure management. The Solution: EdgeSentinel EdgeSentinel is a secure, autonomous, bare-metal Linux Site Reliability Engineering (SRE) Copilot. It fundamentally shifts edge diagnostics from reactive "guess and check" to intelligent, verifiable incident response, powered entirely by local AI. EdgeSentinel in Action Unlike naive scripts that crash upon hitting missing dependencies (like nmcli), EdgeSentinel is a capability-aware platform. It dynamically maps host capabilities at startup, seamlessly falling back to portable alternatives like ip link or reading /sys/class/net directly from the kernel if needed. It never dumps raw bash errors; instead, it intercepts failures and normalizes data through strict Zod-validated contracts. Operators instantly see deterministic states (e.g., "Degraded") on a custom Cyberpunk-styled React telemetry HUD, complete with exact hardware observations and safe remediation steps. Enterprise-Grade Security & Zero-Trust Powered by the Model Context Protocol (MCP) and local LLMs via Ollama, EdgeSentinel ensures zero sensitive system telemetry ever leaves the physical device. It operates on a strict human-in-the-loop safety model: it is read-only by default, and mutating actions (like restarting a service) require explicit human approval after a read-only preflight check. Conclusion: EdgeSentinel is an "SRE in a box". It neutralizes command injection risks, eliminates hallucination-inducing OS errors, and bridges the gap between low-level kernel diagnostics and agentic AI to build unbreakable edge infrastructure.