Loading ToolBench
MCP server for clinical trial design. Computes sample size, power, group-sequential boundaries, NPH (MaxCombo / RMST / milestone), Monte Carlo verification, and markdown reporting via the gsDesign / gsDesign2 R packages. Covers Phase 2 and Phase 3 confirmatory designs.
Clinical-trial-design provides 9 well-scoped tools with complete input schemas and descriptions. Tool names follow verb_noun convention (design_*, validate_*, verify_*, generate). Schemas are properly typed with enums for constrained inputs (design_class, comparison, model, strategy). However, descriptions are domain-specific but lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Parameter descriptions are present but terse (avg ~40 chars vs baseline 72). Output schemas are not documented, agents cannot predict response structure. Error handling exists (DesignrToolError) but lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint). Tools are read-only (no state mutation), which is appropriate for statistical design, but this is not declared.
Computes sample size for binary endpoint clinical trials with fixed or group-sequential designs, supporting superiority and non-inferiority comparisons.
Computes sample size for co-primary endpoint designs with hierarchical or parallel testing strategies, supporting multiple endpoints with different effect sizes.
Computes sample size for continuous endpoint clinical trials with fixed or group-sequential designs.
Computes sample size for designs with graphical multiple comparison procedures (Maurer-Bretz method), supporting gating and hypothesis dependencies.
Computes sample size for multi-population (subgroup) designs with nested or independent populations, supporting hierarchical testing strategies.
Generates a markdown report summarizing a clinical trial design, including sample size, power, boundaries, and reasoning chain.
Output schemas not documented. Agents cannot predict response structure (sample size, boundaries, power, reasoning chain). Forces agents to guess field names for downstream processing.
Tool descriptions lack WHEN-to-use guidance. 'Computes sample size for binary endpoint...' states WHAT but not WHEN to call design_binary vs design_continuous vs design_survival. LLMs must infer from context.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only and idempotent, but this is not declared. Agents cannot optimize retry logic or understand safety guarantees.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
Computes sample size and events for survival endpoint clinical trials with proportional hazards (PH) or non-proportional hazards (NPH) models (MaxCombo, RMST, milestone), supporting fixed and group-sequential designs.
Validates a design against published clinical trial benchmarks, comparing computed sample size and design parameters against historical precedents.
Performs Monte Carlo simulation to verify that a computed design achieves the specified power and Type I error rate under the assumed effect size.
Error handling returns DesignrToolError with class and field, but lacks recovery guidance. 'designr error [InvalidInput] on field alpha: must be in (0, 1)' is clear, but no suggestion for next step (e.g., 'Try alpha=0.05').
Parameter descriptions are terse (avg ~40 chars vs baseline 72). E.g., 'Control arm event rate (0-1)' lacks context on what 'event' means in the trial domain or when to use vs other endpoints.