Hardware-Probe: Deep Hardware Diagnosis and LLM Optimization Tool for AI and High-Performance Computing

An MCP protocol server that provides deep system insights beyond simple spec sheets, designed specifically for AI inference, gaming, and high-performance computing scenarios. It supports real-time performance monitoring, thermal diagnostics, and local LLM runtime optimization.

In AI local inference, gaming, and high-performance computing scenarios, hardware performance bottlenecks are often hidden beneath surface specifications. Users often face confusion like: Why is my high-end graphics card not running LLM at ideal speed? Why does the system slow down for no apparent reason? Traditional system monitoring tools only provide surface-level information, making it difficult to diagnose the root cause of real issues.

yamaru-eu/hardware-probe project emerged as a solution. It is an expert-level hardware probing and performance diagnosis engine built on the Model Context Protocol (MCP), aiming to provide developers and advanced users with deep system insights beyond simple spec sheets.

The project can comprehensively analyze key components of the system:

• CPU Analysis: Detailed detection of processor model, core count, frequency, and architectural features
• Memory Diagnosis: RAM capacity, frequency, channel configuration, latency parameters
• GPU Deep Detection: Not only identifies graphics card model but also deeply analyzes VRAM capacity, memory bandwidth, CUDA core count/stream processor quantity
• Storage Topology: Disk type, interface speed, SMART health status
• OS Environment: Driver versions, runtime libraries, system configurations

Unlike static hardware information collection, hardware-probe supports dynamic system load monitoring:

• Real-time tracking of CPU, GPU, and memory usage changes
• Identifies processes with the highest resource consumption
• Detects I/O bottlenecks and storage performance degradation
• Analyzes memory pressure and Resident Set Size (RSS)

This is one of the tool's most distinctive features. Many users' "mysterious performance drop" issues often stem from thermal throttling:

• Real-time monitoring of CPU/GPU temperature status
• Detects frequency clipping phenomena
• Analyzes fan speed and heat dissipation efficiency
• Identifies performance loss caused by overheating

For the currently popular local Large Language Model (LLM) inference scenarios, hardware-probe provides specialized optimization tools:

• LLM Compatibility Detection: Predicts the running performance of specific models on current hardware
• Quantization Adaptation Calculation: Helps users determine the optimal model quantization scheme (e.g., 4-bit, 8-bit)
• Runtime Optimization Recommendations: Configuration tuning for different inference frameworks like Ollama, CUDA, Metal
• Inference Configuration Analysis: Deeply checks AI runtime environment variables and configuration parameters

hardware-probe uses the Model Context Protocol (MCP) as the underlying communication protocol, meaning it can seamlessly integrate into MCP-supported AI assistants and development tools. Currently, official support includes:

• Gemini CLI: One-click installation via gemini extension install @yamaru-eu/hardware-probe
• Claude Desktop: Usable by configuring MCP server settings
• Other MCP-compatible tools: Access via standard MCP configuration

The project exposes multiple powerful tool interfaces for AI assistants to call:

Tool Name	Function Description
analyze_local_system	Perform a complete hardware inventory scan
analyze_performance	Get real-time performance metrics and top processes
analyze_ram_pressure	Deep memory pressure and RSS analysis
check_storage_health	Disk SMART health check and I/O bottleneck analysis
thermal_profile	CPU/GPU thermal status, fan speed, and frequency throttling detection
diagnose_antivirus_impact	Detect EDR/antivirus software conflicts and development path exclusion coverage
monitor_system_health	Statistical health report over a specified duration (min/max/average values)
check_llm_compatibility	Predict performance of specific LLM models (Beta)
get_llm_recommendations	Recommend models best suited for local execution (Beta)
analyze_inference_config	Deep analysis of AI runtime and configuration environment