Neural Mcp

io.github.andylbrummerPublished by GitHub account andylbrummerv0.1.6Updated Oct 24, 2026

GPU-accelerated MCP server for neural network training, deep learning, and model experimentation

VerifiedSTDIODesktop onlyData & AnalyticsAI & ML

Overview

AI-generated overview

Local MCP server that lets an assistant build, train, evaluate and export neural network models, with optional CUDA acceleration.

What you can do
  • Define custom architectures or load pretrained ResNet and MobileNet models, summarize layers, and export to ONNX or TorchScript
  • Load standard datasets such as MNIST, CIFAR-10 and ImageNet and build batched dataloaders
  • Run training loops with configurable learning rates and batch sizes, monitor experiment status, and evaluate on test sets
  • Compute metrics, generate confusion matrices, plot training curves and visualize predictions
  • Run automated hyperparameter searches
When to use it
Useful when you want an assistant to drive deep learning experiments on your own machine: quick model prototyping, training runs on standard datasets, transfer learning, evaluation and hyperparameter search, without writing the PyTorch plumbing yourself.
Requirements
Runs as a local stdio process, launched with uvx or installed via pip from the scicomp-neural-mcp package; Python and PyTorch are implied. GPU acceleration is opt-in through the MCP_USE_GPU environment variable. Desktop only; no authentication or API keys declared.
Before you install
Training and hyperparameter tuning are long, compute-heavy jobs that can occupy local CPU or GPU for extended periods. Datasets and pretrained weights are downloaded from torchvision or Hugging Face, and models are written to disk when exported. No credentials are requested.

Installation

In SourceWeft

  1. Open Neural Mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

scicomp-neural-mcp

mcp-name: io.github.andylbrummer/neural-mcp

MCP server for neural network training and experimentation.

Overview

This server provides tools for building, training, and analyzing neural networks:

  • Model building - Pre-trained models (ResNet, MobileNet) and custom architectures
  • Dataset loading - CIFAR-10, MNIST, ImageNet datasets with standard preprocessing
  • Training - Full training loops with configurable learning rates and batch sizes
  • Evaluation - Model evaluation, metrics computation, and analysis
  • Hyperparameter tuning - Automated hyperparameter search
  • Export - Export models to ONNX and TorchScript formats
  • GPU acceleration - Optional CUDA acceleration for training

Installation & Usage

bash
# Run directly with uvx (no installation required)uvx scicomp-neural-mcp
# Or install with pippip install scicomp-neural-mcp
# With GPU support (recommended for training)pip install scicomp-neural-mcp[gpu]
# Run as commandscicomp-neural-mcp

Available Tools

Model Management

  • define_model - Create neural network models (ResNet18, MobileNet, custom)
  • load_pretrained - Load pretrained models from torchvision or Hugging Face
  • get_model_summary - Get detailed layer-by-layer breakdown
  • export_model - Export to ONNX or TorchScript

Data Loading

  • load_dataset - Load standard datasets (CIFAR-10, MNIST, ImageNet)
  • create_dataloader - Create batched dataloaders with shuffling

Training

  • train_model - Train model on dataset with configurable parameters
  • get_experiment_status - Monitor training progress
  • evaluate_model - Evaluate on test set

Analysis

  • compute_metrics - Compute detailed performance metrics
  • confusion_matrix - Generate confusion matrices
  • plot_training_curves - Visualize loss and accuracy curves
  • visualize_predictions - Inspect model predictions on samples

Hyperparameter Optimization

  • tune_hyperparameters - Automated hyperparameter search

Configuration

Enable GPU acceleration with environment variable:

bash
MCP_USE_GPU=1 scicomp-neural-mcp

Examples

📖 Code Examples

Practical tutorials in EXAMPLES.md:

  • MNIST digit recognition (complete workflow)
  • Transfer learning with ResNet
  • Hyperparameter optimization
  • Confusion matrix analysis
  • Progressive learning path (beginner → advanced)

📚 Full Documentation

See the API documentation for complete API reference.

Part of Math-Physics-ML MCP System

Part of a comprehensive system for scientific computing. See the documentation for the complete ecosystem.

Source: servers/neural-mcp/README.md at commit 798e264

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.6LatestSep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts