DrugCore and RegulatoryCore are now live in the API and MCP. Read more

Amass
MCP showcasesTirzepatide evidence brief

MCP Showcase

Tirzepatide evidence brief

Name a drug asset; get a citation-auditable Word evidence brief — pivotal Phase-3 trials, grounded efficacy and safety findings, and the trial↔paper links — every figure verbatim from Amass.

See the prompt

New to Amass? Connect Amass to Claude first

5Phase-3 trials
4efficacy claims
1safety signal
You getdocx1 file
Promptregulatoryclaude

The prompt

Paste this into Claude

Drug assettirzepatide
prompt
DRUG ASSET: tirzepatide

You have the Amass MCP connected (TrialCore + BioMedCore). Assemble a one-document EVIDENCE
BRIEF for this asset — citation-auditable, every figure traceable to an Amass record.

1. PIVOTAL TRIALS. Call search_amass_trialcore_records with:
     query = the DRUG ASSET, phase = "PHASE3", interventionType = "DRUG"
   Keep the registrational / outcomes trials. Build a table: NCT | Acronym | Indication |
   Sponsor | Status | Enrollment | Start — every field verbatim from the record.

2. EFFICACY EVIDENCE. Call search_amass_biomedcore_records with:
     query = the DRUG ASSET + key indication terms, minJournalQualityJufo = 2, isRetracted = false
   For each pivotal finding write one CLAIM grounded in a specific abstract. Quote the numbers
   (%, kg, HR, CI, P) VERBATIM from that abstract — never rounded, never recalled. Tag each claim
   with PMID, journal, JuFo tier, citationCount.

3. CROSS-CORE PROVENANCE. For one pivotal trial, call get_amass_trialcore_record and read its
   referencesBiomedCore links — show the trial record pointing at its own describing paper
   (NCT ↔ PMID). This is the publication↔trial graph no raw registry exposes.

4. SAFETY. Include at least one safety signal from a JuFo>=2 abstract, with its verbatim effect
   size — do not soften or omit it.

5. Assemble sections 1–4 as a formatted .docx evidence brief. Close with an AUDIT TRAIL listing
   every Amass ID and PMID cited, and a scope note: figures are verbatim from Amass records as of
   <date>; this is an evidence digest, not regulatory advice.

Why not just ask the model?

Ask a plain model for "a tirzepatide evidence brief" and it produces a fluent document with invented trial IDs, mis-remembered effect sizes, and citations that don't resolve.

  • Plain model: plausible NCTs, recalled (often wrong) effect sizes, unresolvable citations.
  • Amass: real TrialCore records (every NCT, sponsor, status and enrollment verbatim) plus a JuFo-gated, non-retracted corpus — the −20.8-point weight difference (PMID 37840095), the −5 mmHg systolic drop (PMID 39551891), the 2.15 retinopathy odds ratio (PMID 40637847) — each copied from the abstract, and the SUMMIT trial linked straight to its paper (NCT04847557 → PMID 39551891).

Honest scope

An evidence digest, not a label, a systematic review, or regulatory advice. The trial table is the Phase-3 drug set Amass returns for the asset, not an exhaustive program census; the efficacy and safety claims are a grounded sample of the high-quality literature, not a meta-analysis. Every figure is verbatim from an Amass record as of the run date and traceable by the PMIDs and Amass IDs in the audit trail.

Run it yourself

Connect Amass to Claude and paste the prompt.

No build, no deploy — the connector takes about a minute, then this workflow runs on your own inputs.