convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
david-morgenstern

HWInfo MCP Server

david-morgenstern/hwinfo-mcp

Acesso aos dados de hardware do HWiNFO direto do Claude em Windows

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que expõe dados de hardware em tempo real do HWiNFO para Claude em PCs Windows. Fornece timings de memória, relatório completo de hardware (CPU, placa-mãe, GPU, drives) e leitura de sensores ao vivo via interface de memória compartilhada.

Instalação

$npx convez-ai david-morgenstern/hwinfo-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
v0.1.0
Issues abertas
0
Contribuidores
1

Casos de uso

  • Monitoramento de sensores de hardware em tempo real durante tarefas do Claude
  • Diagnóstico de timings de memória e configurações SPD
  • Análise de relatórios de hardware completos de máquinas Windows
  • Automação de coleta de dados técnicos de hardware para troubleshooting

Veredito

Ferramenta muito especializada para técnicos que precisam integrar dados brutos de hardware do Windows ao Claude. Bem implementada com fallbacks para edição gratuita do HWiNFO, mas exige setup manual em cada máquina Windows.

README

A tradução para o português deste README ainda está na fila.

hwinfo-mcp

An MCP server that exposes HWiNFO data on a Windows PC to Claude:

  • Memory timings + SPD (tCAS/tCL, tRCD, tRP, tRAS, tRC, Command Rate, speed) — like the values HWiNFO64 shows in its memory report.
  • Full hardware report (CPU, motherboard, memory, GPU, drives), queryable by section.
  • Live sensors (temperatures, clocks, voltages, fans, usage, power) via HWiNFO's Shared Memory interface.

It runs on the Windows machine (where HWiNFO and the hardware live) and talks to Claude over stdio. Built it on a Mac — copy this folder to the Windows PC to use it.


How it works

DataSource
Live sensorsHWiNFO Shared Memory v2 (Global\HWiNFO_SENS_SM2), read with stdlib mmap — no native deps
Memory timings / full reportHWiNFO report file: tries HWiNFO64.exe -j<file> (JSON), falls back to a manually-saved report

Memory timings are static config data, so they only appear in HWiNFO reports, not in the live shared-memory sensor stream — hence the two paths.

Free-edition caveats ⚠️

You said the edition is free / unknown. Two things to know:

  1. CLI report automation may be Pro-only. If get_memory_timings / get_hardware_report report that no file was produced, use the GUI fallback: in HWiNFO do File → Save Report (pick JSON) into a folder, then set HWINFO_REPORT_DIR to that folder. The server will parse the newest report there.
  2. Shared Memory on the free edition stops after ~12 hours and needs HWiNFO restarted. Enable it under HWiNFO → Settings → Main Settings → "Shared Memory Support", and keep HWiNFO running.

Tools exposed

  • get_memory_timings() — RAM timings + SPD from the report.
  • get_hardware_report(section?) — full report, or just subtrees matching section (e.g. "memory", "cpu").
  • get_live_sensors(filter?) — all live readings, optionally filtered (e.g. "gpu", "fan").
  • get_memory_sensors() — live readings related to memory.
  • get_raw_report(fmt?) — raw report text (json/text/csv/xml) to inspect the real schema.
  • hwinfo_status() — diagnostics (exe found? shared memory readable? reading count).

Install on the Windows PC

Repo: https://github.com/david-morgenstern/hwinfo-mcp · prebuilt assets on the Releases page.

A. pip straight from GitHub (needs Python 3.10+) — simplest

pip install "git+https://github.com/david-morgenstern/hwinfo-mcp"

Installs a hwinfo-mcp command.

B. Standalone .exe — no Python needed

Download hwinfo-mcp.exe from the latest release and put it anywhere, e.g. C:\Tools\hwinfo-mcp.exe.

C. pip from a downloaded wheel

Download hwinfo_mcp-*.whl from the release, then:

pip install hwinfo_mcp-0.1.0-py3-none-any.whl

D. From a local checkout

cd hwinfo-mcp
pip install .        # or: pipx install .

The release .exe is produced by the release GitHub Actions workflow (it also builds the wheel/sdist). Cut a new release by pushing a tag: git tag v0.1.1 && git push origin v0.1.1.

Register with Claude Code

:: if installed via pip/pipx:
claude mcp add hwinfo -- hwinfo-mcp

:: if using the standalone exe:
claude mcp add hwinfo -- "C:\path\to\dist\hwinfo-mcp.exe"

With environment overrides (example):

claude mcp add hwinfo ^
  -e HWINFO_EXE="C:\Program Files\HWiNFO64\HWiNFO64.exe" ^
  -e HWINFO_REPORT_DIR="C:\Users\you\Documents\hwinfo-reports" ^
  -- hwinfo-mcp

For Claude Desktop, add to claude_desktop_config.json:

{
  "mcpServers": {
    "hwinfo": {
      "command": "hwinfo-mcp",
      "env": {
        "HWINFO_EXE": "C:\\Program Files\\HWiNFO64\\HWiNFO64.exe",
        "HWINFO_REPORT_DIR": "C:\\Users\\you\\Documents\\hwinfo-reports"
      }
    }
  }
}

Configuration (env vars)

VarMeaningDefault
HWINFO_EXEFull path to HWiNFO64.exesearched in common install dirs + PATH
HWINFO_REPORT_DIRFolder of manually-saved reports (fallback)none
HWINFO_REPORT_TIMEOUTSeconds to wait for a generated report45

Verifying on Windows

  1. Start HWiNFO (sensors running, Shared Memory enabled).
  2. claude mcp add hwinfo -- hwinfo-mcp, then in Claude ask it to call hwinfo_status.
  3. Ask for get_live_sensors filter="cpu" and get_memory_timings.

Status / honesty

This package was written on macOS and has not been run against real HWiNFO/hardware yet — there's no Windows box in this environment. The shared-memory decoder follows HWiNFO's published v2 layout and is padding-tolerant; the report parser is schema-tolerant/best-effort because HWiNFO's exact JSON key names vary by version. If get_memory_timings looks thin on your machine, run get_raw_report and the field-name heuristics in report.py (_TIMING_KEYS) can be tightened to match.

Ver o original no GitHub

Bom para

Relacionados