AmericanPeptide.com

An AI-grounded research reference for peptides — a trust layer, not a store.

Browse a curated catalog of 56 research peptides, ask a Claude-powered agent that grounds every answer in PubChem, ClinicalTrials.gov, PubMed & UniProt, decode a Certificate of Analysis, and run bench calculators offline. All content is research-framed and openly licensed (CC BY 4.0) so it stays useful wherever it travels.

Live site · Research agent · Remote MCP · Developers

[!IMPORTANT] Research and educational reference only — not a medical device, not medical advice, and not a store. Every AI output is a computational hypothesis requiring validation. Peptide profiles, dosing reference figures, and mechanisms are framed for laboratory research, not human use. User-facing pages carry this disclaimer and a research-use gate.

---

What it is

AmericanPeptide.com treats the domain as a trust and authority asset and builds a research layer on top of a structured data layer:

• Peptide catalog — research peptides with mechanisms, key research, storage/handling, synthesis notes, molecular data, and per-compound FAQs. Ships as a research reference, with no fabricated supplier or price data; sourcing lives in the separate trust-ranked vendor directory.
• Peptide Agent — a Claude research agent that thinks adaptively and grounds claims in live scientific sources rather than model memory.
• Remote MCP server — connect the whole catalog + grounding tools into Claude, ChatGPT, Cursor, or any MCP client.
• Bench tools — reconstitution & blend calculators, a COA decoder, and a design lab, installable as an offline PWA for use at the bench.

Feature tour

Area	Highlights
Catalog	/catalog with category & comparison views, /catalog/[slug] profiles, side-by-side /compare/[pair], glossary, research areas, and synthesis walkthroughs.
Peptide Agent (/research)	Anthropic Claude with adaptive extended thinking, a grounding toolset, streamed responses, and multi-model failover. Cited answers with canonical links.
Reconstitution calculator	Vial/dose/water solver on a U-100 syringe with a live diagram, doses-per-vial, a reference matrix, peptide-aware autofill presets, mg-scale presets, and shareable URLs. Works offline.
COA decoder (/tools/coa-decoder)	Parses a Certificate of Analysis (incl. PDF via pdfjs-dist) and explains what the numbers mean.
Design lab & blend calculator	Compound/blend builders under /tools and /compounds.
Developer surface	Public catalog API, issued API keys with usage metering, and a Streamable-HTTP MCP endpoint — all sharing one data + grounding core.
Offline / PWA	Service worker + offline reference so the bench tools keep working with no network.

How the Peptide Agent is grounded

The agent (src/lib/agent-tools.ts, served at /api/chat) can call free, key-less scientific APIs and cite them — so answers are anchored to primary data:

Tool	Source
search_pubchem	PubChem — chemical identity, mass, formula
search_uniprot	UniProt — protein/sequence data
search_clinical_trials	ClinicalTrials.gov v2 — trial status & endpoints
search_pubmed	PubMed E-utilities — literature

Model selection and an ordered multi-model failover chain live in src/lib/models.ts, so a single upstream outage doesn't take the agent down. The active model is configurable at runtime via the AGENT_MODEL environment variable.

Remote MCP server

Connect the catalog as a live data source from any MCP client (Streamable HTTP transport):

https://americanpeptide.com/api/mcp

Its tools reuse the same catalog data and grounding executors as the on-site agent, and every result carries canonical URLs + CC BY 4.0 attribution so content that travels keeps pointing home.

Tech stack

• Framework: Next.js 15.5 (App Router, Turbopack) · React 19 · TypeScript (strict)
• Styling: Tailwind CSS v4 · Lucide icons · Recharts
• AI: Anthropic Claude via raw fetch (no SDK); grounding tools; static system prompts sent as cached blocks
• MCP: mcp-handler (Streamable HTTP)
• Storage: Upstash Redis (REST) for rate limiting, API keys & usage metering — with in-memory fallback for local dev
• Analytics: Vercel Analytics + Speed Insights, optional GA4
• Docs/PDF: react-markdown + remark-gfm · pdfjs-dist

Quick start

Prerequisites: Node.js 20+ and npm.

git clone https://github.com/utilityincoming/american-peptide.git
cd american-peptide
npm install

# Configure environment
cp .env.example .env.local   # then fill in values (see below)

# Run the dev server (Turbopack)
npm run dev

Open http://localhost:3000.

The site runs with zero config for browsing and the bench tools. The Peptide Agent needs ANTHROPIC_API_KEY; everything else (Redis, GA) is optional and degrades gracefully in local dev.

Environment variables

Full documentation lives in .env.example. Summary:

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗