Loading ToolBench
AI-powered patient advocacy tool for cancer treatment management. Searches PubMed and ClinicalTrials.gov for relevant research, tracks treatment events, provides lab trend analysis, and manages patient data through the Oncofiles MCP server.
Oncoteam has 11 tools with clear verb-noun naming (get_*, search_*, check_*) and domain-appropriate descriptions (avg ~120 chars). All tools have descriptions and most parameters are typed. However, critical gaps exist: input schemas are not visible in the provided source code for most tools, parameter descriptions are sparse or missing, output schemas are undocumented, and error handling guidance is absent. The code snippet shows tool registration via FastMCP but does not expose the actual schema definitions. Without visible schemas, per-tool scores are capped at 50 for inference. The domain (oncology, clinical trials, patient safety) demands high precision, missing parameter constraints and output documentation create risk.
Verify patient eligibility for a specific clinical trial based on biomarkers and clinical criteria
Retrieve AI-generated briefings with research summaries, clinical trial matches, and action items for the patient
Retrieve lab results with trend analysis, SII calculation, and clinical alerts. Returns CBC delta table, liver enzyme patterns, and thrombosis flags.
Retrieve current medication schedule and dosing information for the patient
Retrieve patient biomarkers, excluded therapies, active therapies, and genetic profile for treatment safety checks
Retrieve comprehensive patient profile including diagnosis, staging, treatment regimen, ECOG status, and biomarkers
Input schemas not visible in source code. Only 2 of 11 tools show parameter definitions (search_pubmed, search_clinical_trials with query/limit/status). Remaining 9 tools have no visible schema, cannot verify type constraints, required fields, or parameter descriptions. Per HARD SCORING RULES, schema score must be 0 for tools without visible schemas.
Output schemas completely undocumented. No tool description specifies what fields are returned, their types, or structure. For a clinical tool returning lab results, biomarkers, or trial eligibility, LLMs cannot plan downstream calls or extract safety-critical data without knowing the response shape.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Retrieve patient treatment timeline including chemotherapy cycles, lab results, consultations, imaging, and surgery events
Log patient activity, tool calls, and clinical decisions to audit trail for compliance and monitoring
Search ClinicalTrials.gov for relevant trials matching patient biomarkers, diagnosis, and eligibility criteria
Search PubMed for relevant oncology research articles based on patient biomarkers and diagnosis
Read and extract content from pathology, genetics, or clinical documents. Always call this for biomarker extraction from documents.
Parameter descriptions missing or minimal. view_document has only 'Document identifier' for document_id. get_labs has only 'Number of recent lab entries to retrieve' for limit, no guidance on valid range, default, or what 'recent' means. log_activity has only 'Type of activity' and 'Activity details', no enum of valid activity_type values or schema for details object.
No error handling guidance. Descriptions do not explain what happens on failure (e.g., 'If document not found, returns empty result' or 'If patient ineligible, returns reason codes'). Agents cannot distinguish retryable errors from user-fixable ones, risking silent failures in clinical workflows.
No pagination or result limits documented. search_pubmed and search_clinical_trials accept 'limit' but no description of max value, default, or whether results are sorted by relevance. get_labs accepts 'limit' but no guidance on whether it returns oldest-first or newest-first. Large result sets risk context window exhaustion.