Math-Physics-ML MCP System

Welcome to the Math-Physics-ML MCP System documentation!

The Math-Physics-ML MCP System is a comprehensive, GPU-accelerated platform providing Model Context Protocol (MCP) servers for computational mathematics, physics simulations, and machine learning.

Overview

This system consists of 4 specialized MCP servers that work together to provide a complete computational science platform:

🧮 Math MCP

Symbolic algebra (SymPy) and GPU-accelerated numerical computing

• 14 tools for symbolic math, linear algebra, FFT, and optimization
• Foundation layer used by other MCPs

⚛️ Quantum MCP

Wave mechanics and Schrödinger equation simulations

• 12 tools for quantum simulations
• Split-step Fourier solver for time-dependent wave evolution
• Support for 1D and 2D systems

🔬 Molecular MCP

Classical molecular dynamics simulations

• 15 tools for particle systems and MD simulations
• NVE, NVT, and NPT ensemble support
• Analysis tools for RDF, MSD, and phase transitions

🧠 Neural MCP

Neural network training and experimentation

• 16 tools for deep learning workflows
• Pre-built architectures (ResNet, MobileNet) and custom models
• Hyperparameter tuning and evaluation

Key Features

• GPU Acceleration: Automatic CUDA detection with graceful CPU fallback (10-100x speedup)
• Async Tasks: Long-running operations use MCP Tasks primitive
• Token Efficiency: URI-based references for large arrays and simulation data
• Cross-MCP Workflows: Servers can exchange data and build on each other
• Progressive Discovery: Built-in info tools for capability exploration
• Comprehensive Testing: 75+ tests covering all functionality

Quick Start

Get started with the Math-Physics-ML MCP system:

1. Installation - Set up the system and dependencies
2. Configuration - Configure the MCP servers
3. Quick Start Guide - Run your first computations

Architecture

The system is built as a monorepo with shared infrastructure:

math-mcp/
├── servers/
│   ├── math-mcp/          # Symbolic & numerical computing
│   ├── quantum-mcp/       # Quantum mechanics
│   ├── molecular-mcp/     # Molecular dynamics
│   └── neural-mcp/        # Machine learning
├── shared/
│   ├── mcp-common/        # GPU manager, async tasks, config
│   └── compute-core/      # Unified array interface, FFT, linalg

Learn more in the Architecture guide.

API Reference

Explore the complete API documentation for each MCP server:

• Math MCP API - Symbolic algebra and numerical computing
• Quantum MCP API - Wave mechanics and simulations
• Molecular MCP API - Molecular dynamics
• Neural MCP API - Neural network training

Example: Quantum Wave Scattering

Here's a simple example combining Math MCP and Quantum MCP:

# Create a Gaussian potential barrier (Math MCP)
potential = create_array(
    shape=[256],
    fill_type='function',
    function='10*exp(-(x-128)**2/100)'
)

# Create a Gaussian wave packet (Quantum MCP)
wavepacket = create_gaussian_wavepacket(
    grid_size=[256],
    position=[64],
    momentum=[2.0],
    width=5.0
)

# Solve Schrödinger equation (Quantum MCP)
simulation = solve_schrodinger(
    potential=potential['array_id'],
    initial_state=wavepacket,
    time_steps=1000,
    dt=0.1,
    use_gpu=True
)

Performance

The GPU acceleration provides significant speedups:

MCP	Operation	CPU Time	GPU Time	Speedup
Math	Matrix multiply (1000×1000)	~100ms	~1ms	100x
Quantum	1D Schrödinger (1000 steps, 256 grid)	~30s	~5s	6x
Molecular	MD (100k steps)	~minutes	~seconds	>10x

Status

All 4 MCP servers are fully implemented and tested:

• ✅ 75/75 tests passing
• ✅ GPU and CPU modes supported
• ✅ Pre-commit hooks configured
• ✅ Ready for production use

License

MIT License - see LICENSE file for details.