hardware-probe — Claude Code MCP server: what it does, how to install
This server feeds system diagnostics, thermal metrics, and storage health directly to your AI agent. It helps tune local model inference and identify hardware bottlenecks.
claude mcp add --env REMOTE_API_URL=<REMOTE_API_URL> --transport stdio hardware-probe -- npx -y @yamaru-eu/hardware-probeWhat hardware-probe does
The server connects to local machine telemetry through the Model Context Protocol. It exposes tools to inspect CPU load, RAM pressure, disk health via SMART metrics, and active thermal states. AI agents can inspect clock throttling and fan speeds whenever system performance drops unexpectedly.
A dedicated toolset targets local AI runtimes like Ollama, CUDA, and Metal. It inspects quantization fit against available VRAM and reviews environment configs. Beta tools that estimate model compatibility query an external endpoint, so you must define the REMOTE_API_URL environment variable for them.
Hardware identifiers are stripped before transmitting any data remotely. The monitoring tool can record health statistics over windows up to 10 minutes. A specialized check also detects conflicts where antivirus scanners inspect active developer directories.
What it helps with
- check CPU and GPU temperatures when builds or tasks slow down
- verify whether a quantized model fits into available VRAM
- track down processes creating excessive memory pressure
- inspect storage SMART attributes and antivirus interference on project paths
Example prompts for the agent
- Run analyze_local_system to inspect current GPU memory and CPU topology
- Check if the system is throttling frequencies due to thermal limits
- Analyze RAM pressure and identify which processes consume the most memory
How to install hardware-probe
Platform CLI
claude mcp add --env REMOTE_API_URL=<REMOTE_API_URL> --transport stdio hardware-probe -- npx -y @yamaru-eu/hardware-probeReplace <…> with your own values
Platform CLI
codex mcp add hardware-probe --env REMOTE_API_URL=<REMOTE_API_URL> -- npx -y @yamaru-eu/hardware-probeReplace <…> with your own values
Config file
{
"mcpServers": {
"hardware-probe": {
"command": "npx",
"args": [
"-y",
"@yamaru-eu/hardware-probe"
],
"env": {
"REMOTE_API_URL": "<REMOTE_API_URL>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"hardware-probe": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@yamaru-eu/hardware-probe"
],
"env": {
"REMOTE_API_URL": "<REMOTE_API_URL>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e REMOTE_API_URL=<REMOTE_API_URL> hardware-probe npx -y @yamaru-eu/hardware-probeReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about hardware-probe
How do I install hardware-probe in Claude Code?
Run claude mcp add hardware-probe -- npx -y @yamaru-eu/hardware-probe for standard local diagnostics. If you need remote LLM compatibility tools, pass the API endpoint: claude mcp add --env REMOTE_API_URL=<REMOTE_API_URL> --transport stdio hardware-probe -- npx -y @yamaru-eu/hardware-probe.
What does hardware-probe do?
It provides system telemetry tools for your coding assistant. The agent can examine thermals, memory pressure, storage SMART status, and runtime environments like CUDA or Metal.
Does hardware-probe share private system data?
Hardware identifiers are anonymized before any data leaves your system. Remote network calls occur only when using the beta tools that verify model compatibility through REMOTE_API_URL.
