Loading ToolBench
Data Science MCP Server — Model training, evaluation, and evolution tools for agentic ML workflows. Integrates with agent-utilities IModelEvolver (CONCEPT:AHE-3.15).
Server has 7 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-80 chars, mostly under 100), but many are generic and lack actionable context. Input schemas are present but incomplete: parameters like 'hyperparameters_json' and 'inputs_json' accept free-form JSON strings rather than structured objects, forcing LLMs to construct JSON manually and inviting malformed input. No output schemas documented. Error handling is absent, no guidance on recovery, retryability, or what went wrong. Tool names are reasonable (fit_model, predict, evaluate_model) but lack verb clarity in some cases (evolve_model_class is vague). No parameter validation rules, enums, or constraints visible. The 'ctx' parameter (MCP Context) appears in all tools but is undocumented and non-standard. Security: no evidence of secret injection, permission gates, or audit logging.
Perform k-fold cross-validation for a model class.
Evaluate a fitted model on a dataset split.
Submit a model to the evolutionary Pareto frontier.
Fit a machine learning model on a dataset and return metrics.
Retrieve the current Pareto frontier of model classes.
Generate predictions using a fitted model.
Rank all registered fitted models by their test R2 score.
Free-form JSON string parameters (hyperparameters_json, inputs_json) force LLMs to construct JSON manually, inviting malformed input and parse errors. No validation or schema constraints visible.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., model_id from fit_model, which is needed by predict and evaluate_model).
Non-standard 'ctx' parameter (MCP Context) appears in all tools but is undocumented. Type is 'Context' (not a JSON Schema primitive), and description is generic. Unclear what this parameter does or when to pass it.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
No error handling or recovery guidance. Tools lack descriptions of failure modes, retryability, or what to do if a model_id is invalid or dataset_name does not exist.
Tool 'evolve_model_class' name is vague. Does it modify an existing model, create a new one, or submit it to a queue? Description ('Submit a model to the evolutionary Pareto frontier') clarifies intent but name alone does not convey action clearly.