MCP Machine Learning Agent
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| train_modelsB | Train Logistic Regression and KNN on the breast cancer dataset and export joblib artifacts. |
| get_sample_inputA | Return one test-set patient as a ready-to-use |
| predict_logistic_regressionB | Predict malignant (0) or benign (1) using the saved Logistic Regression model. |
| predict_knnA | Predict malignant (0) or benign (1) using the saved KNN model. |
| confusion_matrix_metricsA | Compute the confusion matrix plus accuracy/precision/recall for both models on the test set. Precision and recall are for the malignant class. Confusion matrix order is [malignant, benign]; rows = actual, columns = predicted. |
| select_best_modelB | Select the best model using a rule (default: recall, then precision, then accuracy). |
| shap_explain_selectedB | Run SHAP for the selected model: global top features plus a local explanation for one test row. Positive SHAP values push toward malignant. |
| export_parquet_auditB | Export a copy of the Parquet audit log for submission. |
| predict_bothA | Predict with BOTH models in one call and report whether they agree. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Each tool targets a distinct pipeline stage (train, predict, evaluate, select, explain, export), and the two single-model predict tools are clearly differentiated by model name. The only overlap is predict_both, which duplicates the two single-model predict tools as a convenience, but its description makes the boundary clear.
Names are consistently snake_case and mostly verb-first (predict_*, train_models, get_sample_input, select_best_model, export_parquet_audit). Minor deviations exist where actions start with a noun (confusion_matrix_metrics, shap_explain_selected), but they remain readable and predictable.
Nine tools map cleanly onto the ML workflow (train, predict, evaluate, select, explain, export) with no filler. The count is well-scoped for a demo pipeline and each tool earns its place.
The surface covers the full lifecycle: training, single and combined predictions, metrics, model selection, SHAP explainability, sample input, and audit export. Minor gaps include no tool to read/inspect the audit log contents directly and no way to pass arbitrary custom patient input beyond the sampled test row.