Two tools with minimal descriptions (both under 100 chars) and incomplete parameter documentation. Input schemas present but lack parameter descriptions entirely, LLMs cannot infer what 'prob_distributions' or 'prob_distribution' objects should contain. No output schema documented. No error handling guidance. Tool names lack action verbs (calibrate_conformal, predict_set are domain-specific but not verb_noun pattern). No parameter constraints, enums, or validation hints. Baseline: 194 chars avg description, 72 chars avg param annotation, this server is 50% below baseline.

• Add 50-150 char descriptions to each parameter. Example: 'true_labels: List of ground-truth class labels (strings) corresponding to each probability distribution. Must match length of prob_distributions.'
• Expand tool descriptions to 150-250 chars. Example for calibrate_conformal: 'Calibrate the non-conformity threshold using a holdout dataset. Call this first with labeled calibration data to set the coverage guarantee level (1-alpha). Returns the quantile cutoff used by predict_set.'
• Document output schemas in tool descriptions or as a separate schema field. Example: 'Returns {status: string, quantile: float, alpha: float}' for calibrate_conformal.
• Wrap error handling in try-catch blocks in handle_request(). Return structured error responses like {error: 'Mismatched calibration data', guidance: 'Ensure true_labels and prob_distributions have the same length.'} instead of raw exceptions.
• Add parameter constraints: alpha should have minValue=0.01, maxValue=0.99 in schema. prob_distributions should document expected dict structure (e.g., 'Dict[str, float] mapping class labels to probabilities, summing to ~1.0').