Neural Mcp

This is one of four specialized servers in a GPU-accelerated scientific computing suite. It exposes 16 tools for building, training, and evaluating neural networks, including define_model for architectures like ResNet, train_model with GPU support, and export_model to ONNX format. Works with standard datasets like CIFAR10 and integrates with the broader math-mcp ecosystem through URI-based data sharing across quantum, molecular, and symbolic math servers. You'd reach for this when you want Claude to prototype neural networks, run training experiments, or set up model export pipelines without leaving the conversation. Async task support handles long-running training jobs. GPU acceleration delivers real speedups where it matters, but everything falls back to CPU if CUDA isn't available.