// MCP TOOL REFERENCE

SCICOMP NEURAL TOOLS

14 tools from the Scicomp Neural MCP Server, categorised by risk level.

READ TOOLS

compute_metrics Compute detailed performance metrics 2/5 confusion_matrix Generate confusion matrices 2/5 get_experiment_status Monitor training progress 2/5 get_model_summary Get detailed layer-by-layer model breakdown 2/5 load_dataset Load standard datasets (CIFAR-10, MNIST) 2/5 load_pretrained Load pretrained models from registries 3/5 plot_training_curves Visualise loss and accuracy curves 2/5 visualize_predictions Inspect model predictions on samples 2/5

WRITE TOOLS

create_dataloader Create batched dataloaders with shuffling 2/5 define_model Create neural network model architecture 3/5 export_model Export model to ONNX or TorchScript 3/5

EXECUTE TOOLS

evaluate_model Evaluate model on test set 3/5 train_model Train model on dataset 4/5 tune_hyperparameters Automated hyperparameter search 4/5

// FAQ

How many tools does the Scicomp Neural MCP server have? +

The Scicomp Neural MCP server exposes 14 tools across 3 categories: Read, Write, Execute.

How do I enforce policies on Scicomp Neural tools? +

Use Intercept, the open-source MCP proxy. Write YAML rules for each tool — rate limits, argument validation, or deny rules — then run Intercept in front of the Scicomp Neural server.

What risk categories do Scicomp Neural tools fall into? +

Scicomp Neural tools are categorised as Read (8), Write (3), Execute (3). Each category has a recommended default policy.

Enforce policies on Scicomp Neural

Open source. One binary. Zero dependencies.

npx -y @policylayer/intercept

github.com/policylayer/intercept →