Exposes a trained regression machine learning pipeline as MCP tools. Provides prediction, explanation, feature inspection, and metrics tracking capabilities for an ML model.
Server exposes 5 tools with clear verb-noun naming (list_features, get_latest_metrics, get_run_history, predict, explain_prediction). Descriptions are present and moderately detailed (avg ~120 chars), meeting the 10-1024 char baseline. However, critical gaps exist: (1) Input parameters lack descriptions, 'features' dict in predict/explain_prediction has no guidance on format, range, or error handling; (2) Output schemas are completely undocumented, LLMs cannot infer what fields to expect from get_latest_metrics or get_run_history; (3) No error handling guidance, what happens if features are missing, invalid, or out of range? (4) No pagination support for list_features or get_run_history despite potentially large result sets. Tools are read-only (low risk) but lack the structured output documentation required for reliable agent composition.
Explain a prediction via per-feature SHAP contributions. Pass a dict of {feature_name: float_value} for every required feature.
Return performance metrics from the most recent pipeline run.
Return metrics from all previous pipeline runs in chronological order.
Return the feature names required to make a prediction.
Make a regression prediction given feature values. Call list_features() first to see which features are required. Pass a dict of {feature_name: float_value} for every required feature.
Output schemas completely undocumented. LLMs cannot infer what fields get_latest_metrics, get_run_history, predict, or explain_prediction return. This violates pattern:tool and forces agents to guess at response structure.
Input parameter 'features' (in predict and explain_prediction) lacks description. No guidance on format (dict of {name: float}?), required keys, valid ranges, or error handling. Violates pattern:tool-description.
No error handling guidance. What happens if predict() receives invalid features (missing keys, non-numeric values, out-of-range)? No recovery hints for LLM. Violates pattern:recovery-guide.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
No pagination support for list_features or get_run_history. If run history grows large, returning all records wastes tokens and risks context window exhaustion. Violates pattern:paginated-result.
Tool descriptions lack dependency hints. predict() description says 'Call list_features() first' but does not explain why or what to do with the result. Violates pattern:tool-description guidance on discovery tools.