Most MCP implementations read files, call APIs, and query databases. A technical writeup from Ingero proposes something lower-level: connecting AI agents directly to Linux kernel tracepoints - the system monitoring hooks that expose exactly what a computer is doing at any given moment.
MCP (Model Context Protocol) is Anthropic's standard for connecting AI models to external tools and data sources - the same mechanism that lets Claude Code read local files and run terminal commands. The Ingero proposal applies that connection layer to eBPF tracepoints, which are small monitoring programs that run inside the Linux kernel and capture raw system data: which processes are active, what system calls they're making, network packet flows, CPU scheduling decisions. This data normally lives in specialized ops tooling, not in an AI agent's field of view.
The result would be an AI agent watching production system behavior directly, not reasoning from static logs after the fact. Instead of a human pulling metrics, correlating timestamps, and manually building a diagnosis, the agent sees the live data stream and works from it continuously.
This is niche today. It requires Linux, familiarity with eBPF tooling, and the technical work to configure the MCP server. But it's a useful marker for where AI agent applications are heading beyond file-reading and API-calling. The more interesting agent use cases watch systems in real time and reason over what changes. Kernel tracepoints are one of the most direct data sources available for that kind of work.