A Python-based optimization solver library demonstrating exact algorithms for route planning (TSP), A/B testing, anomaly detection, bin packing, and break-even analysis with independent verification.
The server has well-structured tool definitions with clear names, comprehensive descriptions, and documented parameter schemas. All 5 tools follow verb_noun naming conventions (solve_*) and include detailed docstrings explaining their purpose, algorithms, and verification methods. However, there are consistent gaps in parameter-level descriptions, output schema documentation, and error handling guidance. Most parameters lack constraint information (ranges, enum values, or format specifications) that would help LLMs invoke tools correctly. Output schemas are not formally documented in the tool definitions, they must be inferred from code. Error handling is minimal; no tools provide recovery guidance or categorization. The server demonstrates good intent with verification logic and feasibility checks, but falls short of production-grade tool design for agent use.
Solve A/B testing problem using frequentist two-proportion z-test, Bayesian Beta-Binomial model, and bootstrap confidence intervals. Computes statistical significance, effect sizes, power analysis, and provides recommendations. Verification uses Fisher's exact test and chi-squared cross-validation.
Detect anomalies and change points in time series data. Uses z-score rolling window method, IQR-based fence detection, and PELT (Pruned Exact Linear Time) algorithm for change point detection. Computes precision, recall, F1 metrics for anomalies and validates detected change points against true regimes using t-tests.
Solve bin packing (container loading) problem by packing items into bins while minimizing number of bins. Compares First Fit Decreasing (FFD) heuristic with optimal Integer Linear Programming (ILP) solution. Verifies feasibility, capacity constraints, and optimality.
Solve break-even analysis problem. Finds the break-even quantity where Revenue = Cost for a new product. Performs sensitivity analysis on price and variable cost changes using symbolic (SymPy) and numerical (NumPy) methods. Returns closed-form symbolic solution and profit tables.
Solve Traveling Salesman Problem (TSP) exactly using Held-Karp dynamic programming algorithm. Finds the shortest Hamiltonian cycle through n cities (warehouse + delivery stops). Complexity: O(n^2 * 2^n) time, O(n * 2^n) space. Returns exact optimal tour with verification.
Output schemas not formally documented in tool definitions. LLMs cannot see what fields the response will contain, forcing them to guess or proceed blind. Solutions return complex objects (Solution dataclass) with multiple fields (tour, tour_names, objective, constraint_check, etc.) but these are not declared in the MCP schema.
Parameter descriptions lack constraint information. Parameters like 'z_window' (int), 'z_threshold' (float), 'alpha' (float) have descriptions but no documented min/max ranges. 'alternative' (str) in solve_ab_test has a description but should declare valid enum values ('two-sided', 'larger', 'smaller') formally, not just in natural language.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No error handling guidance or recovery instructions. Solvers may fail (e.g., invalid inputs, computational limits) but no error classification or actionable recovery steps are defined. LLMs will not know whether to retry, ask the user, or abandon the plan.
Complex tuple types in parameters (tuple[str, ...], tuple[tuple[float, ...]]) may not serialize or deserialize correctly in JSON-RPC. No validation shown for input correctness (e.g., dist_matrix must be square, must match locations length).
solve_ab_test returns multiple statistical models and metrics (frequentist, Bayesian, bootstrap, Fisher's exact, chi-squared cross-validation) but the tool description does not clarify which result the LLM should use or how to interpret conflicts between methods.
Tool descriptions reference internal verification methods ('verified independently', 'certificate', 'constraint_check') but these are implementation details that belong in hidden logic, not in LLM-facing descriptions. This clutters the description with signal-to-noise issues.