@modelcontextprotocol/server-system-monitor

Collects live CPU, memory, disk, and process-level metrics from the host operating system and exposes them through the Model Context Protocol (MCP) as callable tools. Uses native OS APIs (via Node.js child processes or system libraries) to poll system state at configurable intervals, then serializes metrics into structured JSON payloads that LLM clients can query synchronously or subscribe to via MCP's resource subscription mechanism.

Automatically generates MCP-compliant tool definitions (JSON schemas) for each system metric endpoint, enabling LLM clients to discover and call metric-fetching functions with proper type hints and descriptions. The server introspects available metrics at startup and generates OpenAPI-style schemas that describe input parameters (e.g., process filter, metric type) and output structures, which are then advertised via MCP's tools/list endpoint.

Breaks down system metrics to the individual process level, allowing LLM clients to query CPU, memory, and I/O usage per process, with optional filtering by process name, PID, or resource threshold. Internally uses Node.js child processes to invoke system commands (ps, top, or equivalent) and parses their output into structured process records, then applies filter logic to return only relevant processes.

Allows configuration of how frequently the server collects system metrics (e.g., every 1 second, 5 seconds, or on-demand) and how long metrics are cached before being refreshed. Implements a polling loop that runs at a configurable interval, stores the latest snapshot in memory, and serves cached results to clients until the next poll cycle completes. Configuration is typically provided via environment variables or a config file at server startup.

Implements MCP's resource subscription mechanism to enable clients to subscribe to metric updates and receive push-based notifications when metrics change, rather than polling. The server maintains a list of active subscriptions and pushes updated metric snapshots to subscribed clients at each polling interval or when metrics exceed configured thresholds. Uses MCP's resources/subscribe and resources/updated endpoints to manage subscriptions and deliver updates.

Collects and exposes disk-level metrics including I/O throughput (read/write bytes per second), I/O operations per second (IOPS), disk utilization percentage, and available/used space per filesystem. Internally queries the OS filesystem APIs (via df, iostat, or equivalent) and parses output into structured disk metrics, optionally tracking I/O deltas between polling intervals to compute throughput.

Collects network interface statistics including bytes sent/received, packet counts, error rates, and optionally tracks active network connections (TCP/UDP sockets) with their associated processes. Queries OS network APIs (via ifconfig, netstat, ss, or equivalent) and parses output into structured network metrics, optionally computing throughput deltas between polling intervals.

Provides detailed memory usage breakdown including resident set size (RSS), heap usage, external memory, and optionally distinguishes between different memory types (physical, swap, cached). On Linux, parses /proc/meminfo and /proc/[pid]/status for detailed memory accounting; on other OSes, uses available APIs to approximate breakdown. Exposes both system-wide memory and per-process memory details.

Collects CPU metrics including per-core CPU usage percentage, system-wide CPU average, and load averages (1-minute, 5-minute, 15-minute). Computes CPU usage by reading /proc/stat on Linux or equivalent APIs on other OSes, calculating the delta between consecutive snapshots to derive per-core and aggregate CPU percentages. Exposes both instantaneous CPU usage and historical load averages.