AI-powered FDA drug label intelligence system with multi-agent analysis, RAG-based retrieval, and MCP server integration for querying, comparing, and analyzing clinical drug information.
Nine tools with READ_ONLY risk classification and basic schemas present. However, critical gaps undermine quality: (1) Descriptions are present but generic, most lack WHEN/WHY guidance or recovery hints. (2) Parameter descriptions are minimal or absent (e.g., 'drug_name' in fda_label_lookup has no format guidance; 'section' in query_drug_labels lacks enum constraint despite listing valid values in description text). (3) Output schemas are completely undocumented, no return type definitions visible in source. (4) Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. (5) Tool composition issues: rag_search, query_drug_labels, and fda_label_lookup overlap significantly (all search drug labels), creating LLM disambiguation burden. (6) No idempotency guarantees stated. Baseline: 194 chars avg description (these ~100-150 chars); 72 chars avg param annotation (these ~30-50 chars). Most tools fall in C-D range individually.
Search ClinicalTrials.gov for clinical trials related to a drug or condition. Returns trial metadata: phase, status, endpoints, enrollment, sponsor.
Compare FDA drug labels across multiple medications. Provides side-by-side analysis of safety, dosing, or indications.
Compare specific aspects (safety, dosing, indications) across multiple drugs. Retrieves and structures data from the knowledge base for side-by-side analysis.
Look up the latest FDA drug label directly from the openFDA API. Use when the vector store doesn't have information about a specific drug, or when you need the most current label data.
Get the status of the Clinical Document Intelligence Platform including document count and available tools.
List all drugs currently indexed in the knowledge base.
Output schemas completely undocumented. No return type definitions visible for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does rag_search return citations? what fields?). Violates pattern:tool requirement.
Three tools (rag_search, query_drug_labels, fda_label_lookup) perform nearly identical functions, search FDA drug labels. LLM must reason about subtle differences (vector store vs API, optional filters). Violates pattern:tool composition rule: 'Avoid multiple tools that do the same thing differently.'
Parameter descriptions lack actionable constraints. 'section' in query_drug_labels lists valid values in description text ('contraindications, dosage_and_administration, ...') but should be declared as enum in schema. 'drug_name' lacks format guidance (brand vs generic?). Violates pattern:constrained-input.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Search FDA drug labels for information about medications. Returns citation-grounded answers from official FDA prescribing information.
Search the FDA drug label knowledge base for information about drugs, contraindications, dosing, warnings, adverse reactions, and drug interactions. Returns relevant document chunks with citations.
Retrieve all safety-critical information for a drug: black box warnings, contraindications, REMS requirements, and pregnancy category. Always use this before providing dosing or treatment recommendations.
No error handling guidance. Tools lack recovery hints (e.g., 'If drug not found, try search_drugs() with partial name'). No categorization of errors as retryable vs user-fixable. Violates pattern:recovery-guide.
list_available_drugs and get_platform_status have minimal descriptions (50 chars). Descriptions must explain WHEN to call them and what structure they reveal. Current descriptions do not guide LLM selection. Violates pattern:tool-description baseline (194 chars avg).