linux-mcp
# linux-mcp - Linux MCP Server
A Linux system monitoring server built on the [Model Context Protocol (MCP)](https://modelcontextprotocol.io). Provides real-time system information - CPU, memory, disk, network, processes, Docker, and more - via MCP tools over STDIO transport.
[](https://registry.modelcontextprotocol.io) [](https://glama.ai/mcp/servers/Mohabdo21/linux-mcp)

<p align="center">
<img src="assets/demo.gif" alt="linux-mcp demo" width="100%">
</p>
## Features
- **System** - hostname, OS, kernel, architecture, uptime, load averages
- **CPU** - usage, model, frequency, core counts, temperature sensors
- **Memory** - RAM and swap usage with percentages
- **Disk** - per-partition usage, inodes, mount options, largest files, block devices
- **Network** - interface stats, active connections, listening ports, DNS, ping, IP geolocation & ASN lookup
- **Processes** - running processes sorted by CPU or memory, open file descriptors per process
- **Docker** - containers, images, networks, volumes, disk usage, system info, and container stats
- **Services & automation** - systemd units, service status, user timers, crontab, system cron jobs
- **Security** - active user sessions, failed login detection, SELinux/AppArmor status, firewall/SSH/SUID/world-writable audit with security score
- **Packages** - installed packages and available updates (pacman, dpkg, rpm/dnf, apk)
- **Hardware** - GPU info, PCI/USB bus devices, power/battery analytics
- **Desktop session** - Wayland/X11 protocol, DE identifiers, runtime config
- **Storage health** - RAID status, logrotate configuration, time synchronization, SMART disk health, per-device I/O metrics
- **System health** - comprehensive health assessment with memory, disk, load, and systemd checks
- **/proc diagnostics** - deep /proc inspection: interrupts, softirqs, vmstat, diskstats, filesystems, kernel version, slabinfo
- **Man pages** - system manual pages for any installed command
- **Snapshot** - comprehensive system overview in a single call
- **MCP Resources** - system data also accessible as readable resources
## Prerequisites
- **Go 1.27+** (to build from source)
- **Linux** (the server targets Linux; some tools use Linux-specific paths)
- **Docker** (optional - only needed for Docker tools)
## Installation
### Via MCP Registry
Discover the server on the [MCP Registry](https://registry.modelcontextprotocol.io) and install via your MCP client (VS Code one-click install, or manual config).
### Download pre-built binary
Download the latest binary from the [GitHub Releases](https://github.com/Mohabdo21/linux-mcp/releases) page:
```bash
curl -LO https://github.com/Mohabdo21/linux-mcp/releases/latest/download/linux-mcp
chmod +x linux-mcp
```
A fully static build (no libc dependency) is also available as `linux-mcp_static`.
### Via `go install`
```bash
go install github.com/Mohabdo21/linux-mcp@latest
```
The binary is placed in `$GOPATH/bin/` (typically `~/go/bin/`).
### Build from source
```bash
git clone https://github.com/Mohabdo21/linux-mcp.git
cd linux-mcp
make build
```
The binary is placed at `bin/linux-mcp`.
For a fully static binary (no libc dependency):
```bash
make build-static
```
## Usage
The server communicates over STDIO transport, following the MCP standard. It is designed to be launched by an MCP client (e.g., Claude Desktop, Claude Code, Codex, OpenCode, or any MCP host).
### Running directly
```bash
./bin/linux-mcp
```
This starts the server and listens for MCP requests on STDIN/STDOUT.
### Integration
<details>
<summary><h4>OpenCode</h4></summary>
Add the following to your OpenCode configuration:
```json
{
"mcp": {
"linux-mcp": {
"type": "local",
"command": ["/path/to/linux-mcp"],
"enabled": true
}
}
}
```
</details>
<details>
<summary><h4>Claude Code</h4></summary>
Add the server via the CLI:
```bash
claude mcp add linux-mcp -- /path/to/linux-mcp
```
To scope it to all your projects instead of just the current one:
```bash
claude mcp add linux-mcp --scope user -- /path/to/linux-mcp
```
Or create a `.mcp.json` file in your project root for team-shared configuration:
```json
{
"mcpServers": {
"linux-mcp": {
"type": "stdio",
"command": "/path/to/linux-mcp"
}
}
}
```
Verify the connection inside a Claude Code session with `/mcp`.
</details>
<details>
<summary><h4>Codex</h4></summary>
Add the server via the CLI:
```bash
codex mcp add linux-mcp -- /path/to/linux-mcp
```
Or add it manually to `~/.codex/config.toml` (or `.codex/config.toml` for project-scoped config):
```toml
[mcp_servers.linux-mcp]
command = "/path/to/linux-mcp"
```
Verify the connection with `codex mcp list`.
</details>
### Tools and resources
65 tools and 19 resources covering system, CPU, memory, disk, network, processes, Docker, security, packages, hardware, and more.
**[Full tool and resource reference](docs/tools.md)**
## Configuration
The server can be configured via a JSON file loaded from `~/.config/linux-mcp/config.json` or the `LINUX_MCP_CONFIG` environment variable.
```json
{
"log_level": "info",
"timeouts": {
"get_system_snapshot": "60s",
"ping_host": "5s"
},
"disabled": []
}
```
| Field | Description |
| ----------- | --------------------------------------------------------- |
| `log_level` | One of `debug`, `info`, `warn`, `error` (default: `info`) |
| `timeouts` | Per-tool timeout overrides as Go duration strings |
| `disabled` | List of tool names to disable at startup |
The server also handles **SIGHUP** to reload the configuration file at runtime without restarting.
## MCP Registry
Published on the [MCP Registry](https://registry.modelcontextprotocol.io) as `io.github.Mohabdo21/linux-mcp`.
## License
- [MIT](./LICENSE)
## Dependencies
- [modelcontextprotocol/go-sdk](https://github.com/modelcontextprotocol/go-sdk) - MCP SDK for Go
- [shirou/gopsutil/v4](https://github.com/shirou/gopsutil) - System metrics (CPU, memory, disk, network, processes, sensors)
- [docker/go-sdk](https://github.com/docker/go-sdk) - Docker Engine API client
- [moby/moby/client](https://github.com/moby/moby) - Docker client library
- [ip-api.com](https://ip-api.com) - Free IP geolocation API (used by `get_ip_info`)
TDQS
Scored across 65 tools
Tools cover many overlapping domains (CPU, memory, disk, network, process, Docker, etc.), but descriptions explicitly cross-reference alternatives (e.g., 'for X use Y'), making most selections clear. A few clusters such as process info/tree/fds and disk usage/inode/block devices still require careful reading to avoid misselection.
Almost all tool names use snake_case with a consistent get_<noun> pattern. The only exceptions are ping_host and resolve_dns, which use action verbs instead of get_, a minor deviation but still readable and consistent in casing.
With 65 tools, the surface is far beyond the typical well-scoped range (3-15) and exceeds the 50+ threshold for extreme mismatch. Even for a broad Linux diagnostics server, the set includes overlapping utilities (e.g., multiple process, disk, and network tools) that could be consolidated or scoped more tightly.
The read-only diagnostic surface is exceptionally comprehensive, covering system info, hardware, storage, network, Docker, systemd, security, logs, users, packages, and processes. Minor gaps exist (e.g., no sysctl/kernel parameter read, no partition table detail, no arbitrary file-content inspection), but agents can work around most of these.