The tools that watch over a duration are what separate this from a one-shot reading: an intermittent stall does not show up in a single sample, and monitor_cpu_usage and monitor_io_performance are how you catch it. The two snapshot tools are the sensible opening move — one call brings back the whole picture, and you drill in only where it looks wrong.
A cross-platform system monitor exposing fifteen tools, from single readings to composed snapshots. It goes further than usage percentages: I/O rates per disk, context switches and interrupts, page faults and swap activity, interface speed and MTU. GPU coverage spans NVIDIA with full metrics and Apple Silicon including unified memory and core count.
- CPU usage with per-core statistics, frequency, temperature and processor detail
- GPU information for every detected card, including unified memory on Apple Silicon
- RAM and swap, or the detailed version with buffers, cache, active and inactive memory, page faults and swap activity
- Disk usage across mounted drives, and I/O metrics with read and write rates and per-disk busy time
- Network interface statistics, or the enhanced version with transfer rates, errors, drops, speed and MTU
- Load averages over 1, 5 and 15 minutes with context switches, interrupts and running or blocked process counts
- Top processes by CPU or memory
- CPU and I/O watched over a duration you set, rather than sampled once
- A complete system snapshot, a complete performance snapshot, and the current datetime in ISO format
Python with FastMCP, and the machine you want to read — it reports on the host the server runs on. No account or key.