Quick Start
Get started with the Math-Physics-ML MCP servers in minutes.
These are MCP tools, not Python functions
You don't write code. You describe what you want in natural language, and Claude calls the appropriate tools automatically.
Just start Claude and ask for what you want!
How It Works
claude
> "Analyze this sensor data and find the dominant frequencies"
> "Simulate quantum tunneling through a barrier"
> "Train a neural network on my dataset"
Claude invokes the right tools and returns results—no code required.
What Each Server Does
| Server | What It Provides |
|---|---|
| Math MCP | Symbolic algebra, numerical arrays, FFT, optimization, linear algebra |
| Quantum MCP | Quantum potentials, wavefunctions, Schrödinger equation solving |
| Molecular MCP | Particle systems, molecular dynamics, trajectory analysis |
| Neural MCP | Neural networks, datasets, training, evaluation |
Example: FFT Spectral Analysis
Real problem: Robot accelerometer data shows potential bearing wear. Find the dominant vibration frequencies.
You: "I have sensor data sampled at 1000 Hz. Create an array with these values and find the frequency spectrum: [0.1, 0.3, 0.5, 0.2, ...]"
Claude: [calls create_array with the data, then fft]
→ Returns frequency spectrum showing peaks at 15Hz, 45Hz, and 127Hz
You: "Which frequencies are above the normal threshold of 50Hz?"
Claude: The frequencies above 50Hz are 127Hz and its harmonics.
This may indicate bearing defect frequency - recommend inspection.
This is something Claude alone cannot do—it needs actual numerical computation.
Example: Quantum Tunneling
Real problem: Estimate tunneling probability for an electron through a potential barrier.
You: "Simulate a Gaussian wavepacket with momentum 2.0 hitting a 10-unit barrier from x=50 to x=70. What fraction tunnels through?"
Claude: [calls create_custom_potential + create_gaussian_wavepacket + solve_schrodinger]
→ Returns ~23% transmission probability
You: "At what barrier height would tunneling drop below 1%?"
Claude: [runs parameter sweep]
→ Barrier height of ~15 units would reduce tunneling to <1%
Claude can't simulate quantum mechanics without these tools.
Example: Molecular Dynamics
Real problem: Simulate how a protein folds under thermal fluctuations.
You: "Create a 500-atom system at 310K and run NVT dynamics for 5000 steps"
Claude: [calls create_particles + add_potential + run_nvt]
→ Returns trajectory and analyzes structural properties
You: "Compute the radius of gyration over time"
Claude: [calls analyze_trajectory]
→ Returns time series showing protein compactness
LLMs cannot run molecular dynamics simulations.
Example: Neural Network Training
Real problem: Train a classifier on a custom dataset.
You: "Load my image dataset from /data/images and train a ResNet for 10 epochs"
Claude: [calls load_dataset + define_model + train_model]
→ Returns trained model with training curves
You: "What's the accuracy on the test set?"
Claude: [calls evaluate_model]
→ 94.2% accuracy with confusion matrix showing main error sources
Claude can write PyTorch code, but it often has bugs. MCP tools run actual training correctly.
Available Tools
Math MCP
| Tool | Purpose |
|---|---|
symbolic_solve | Solve equations symbolically |
symbolic_diff | Compute derivatives |
symbolic_integrate | Compute integrals |
create_array | Create numerical arrays |
matrix_multiply | GPU matrix multiplication |
solve_linear_system | Solve Ax = b |
fft / ifft | Fourier transforms |
optimize_function | Find minima |
find_roots | Find equation roots |
Quantum MCP
| Tool | Purpose |
|---|---|
create_lattice_potential | Crystal lattices |
create_custom_potential | Custom potentials |
create_gaussian_wavepacket | Quantum states |
solve_schrodinger | 1D Schrödinger equation |
solve_schrodinger_2d | 2D Schrödinger equation |
render_video | Animate wavefunction |
Molecular MCP
| Tool | Purpose |
|---|---|
create_particles | Initialize N-particle system |
add_potential | Add LJ/Coulomb forces |
run_md | NVE dynamics |
run_nvt | NVT dynamics |
run_npt | NPT dynamics |
compute_rdf | Radial distribution |
render_trajectory | Animate trajectory |
Neural MCP
| Tool | Purpose |
|---|---|
define_model | Create architecture |
load_pretrained | Load pretrained models |
load_dataset | Load CIFAR-10, MNIST |
train_model | Train neural network |
evaluate_model | Test set evaluation |
tune_hyperparameters | Hyperparameter search |
export_model | Export to ONNX |
GPU Acceleration
For large computations, enable GPU:
export MCP_USE_GPU=1
claude
GPU provides 10-100x speedup for:
- Large matrix operations
- 2D quantum simulations
- Long MD trajectories
- Neural network training
Interactive Demos
Watch spectacular visualizations:
- Single-Slit Diffraction - Quantum diffraction
- Double-Slit Interference - Interference patterns
- Galaxy Collision - N-body dynamics
Run demos:
claude -p "Simulate double-slit interference" --allowedTools "mcp__quantum-mcp__*"
Next Steps
- 📖 Examples: EXAMPLES.md in each server folder
- 🎬 Visual Demos: Interactive demonstrations
- 📚 API Reference: Complete documentation