🧬 langchain-pubmed

LangChain.js integration for PubMed API - search biomedical literature and retrieve article metadata.

Features

• 🔍 Search PubMed's 35+ million biomedical literature citations
• 🤖 Ready-to-use LangChain Tool for AI agents
• 📄 Returns structured article metadata (title, abstract, publication date)
• ⚡ Built-in retry logic and rate limit handling
• 🔄 Supports streaming results with async iterators
• 🎯 Converts to LangChain Documents for RAG applications

Installation

npm install langchain-pubmed @langchain/core

Quick Start

import { PubMedTool } from "langchain-pubmed";

const tool = new PubMedTool({
  topKResults: 3,
  email: "[email protected]", // Recommended for better rate limits
});

const result = await tool.invoke("covid-19 vaccine efficacy");
console.log(result);

Environment Variables

Set these environment variables to avoid hardcoding credentials:

export PUBMED_EMAIL="[email protected]"
export PUBMED_API_KEY="your_ncbi_api_key"  # Optional, for higher rate limits

Then use without passing credentials:

const tool = new PubMedTool({ topKResults: 3 });

Usage

1. As a LangChain Tool

import { PubMedTool } from "langchain-pubmed";

const tool = new PubMedTool({
  topKResults: 3,
  email: "[email protected]",
});

// Direct invocation
const result = await tool.invoke("diabetes treatment");
console.log(result);

2. With an AI Agent

import { ChatOpenAI } from "@langchain/openai";
import { createAgent } from "langchain";
import { PubMedTool } from "langchain-pubmed";
import dotenv from "dotenv";

dotenv.config();

const model = new ChatOpenAI({
  model: "gpt-5-nano",
});

const tools = [
  new PubMedTool({
    topKResults: 5,
  }),
];

const agent = createAgent({ model, tools });

const result = await agent.invoke({
  messages: [
    {
      role: "user",
      content: "What are the latest treatments for Alzheimer's disease?",
    },
  ],
});

console.log(result.messages[result.messages.length - 1].content);

3. Direct API Wrapper

import { PubMedAPIWrapper } from "langchain-pubmed";

const pubmed = new PubMedAPIWrapper({
  topKResults: 5,
  email: "[email protected]",
  apiKey: "your_ncbi_api_key", // Optional: for higher rate limits
});

// Get structured metadata
const articles = await pubmed.load("cancer immunotherapy");
articles.forEach((article) => {
  console.log(article.Title);
  console.log(article.Summary);
});

// Get LangChain Documents
const documents = await pubmed.loadDocs("machine learning healthcare");

// Stream results
for await (const article of pubmed.lazyLoad("alzheimer disease")) {
  console.log(article.Title);
}

4. For RAG Applications

import { PubMedAPIWrapper } from "langchain-pubmed";

const pubmed = new PubMedAPIWrapper({
  topKResults: 10,
  email: "[email protected]",
});

// Get LangChain Documents for vector store
const documents = await pubmed.loadDocs("CRISPR gene editing");

// Add to your vector store
// await vectorStore.addDocuments(documents);

Configuration Options

| Option | Type | Default | Description | | -------------------------------------------------- | -------- | -------------------------- | ------------------------------- | | topKResults | number | 3 | Number of results to return | | maxQueryLength | number | 300 | Max query length (chars) | | docContentCharsMax | number | 2000 | Max content length (chars) | | maxRetry | number | 5 | Max retries on rate limit | | sleepTime | number | 200 | Initial retry delay (ms) | | email | string | "[email protected]" | Email for PubMed API | | apiKey | string | "" | NCBI API key (optional) | | Plus all ToolParams from @langchain/core/tools | | | Callbacks, tags, metadata, etc. |

Rate Limits

• Without API key: 3 requests/second
• With API key: 10 requests/second

Get a free API key at: https://www.ncbi.nlm.nih.gov/account/settings/

Development

# Install dependencies
npm install

# Build
npm run build

# Run tests
npm test

# Run e2e tests (calls actual PubMed API)
npm run test:e2e

# Run unit tests
npm run test:unit

# Lint
npm run lint

# Format
npm run format

Example

See example.ts for a complete working example.

# Build first, then run the example
npm run build
npx ts-node example.ts

API

PubMedTool

Main tool class for LangChain agents.

const tool = new PubMedTool(options);
const result = await tool.invoke(query);

PubMedAPIWrapper

Core API wrapper with multiple access methods.

Methods:

• run(query) - Get formatted search results string
• load(query) - Get array of article metadata
• loadDocs(query) - Get array of LangChain Documents
• lazyLoad(query) - Async iterator over article metadata
• lazyLoadDocs(query) - Async iterator over Documents

License

MIT

Credits

TypeScript port of the Python LangChain PubMed integration.

Resources

• PubMed
• NCBI E-utilities API
• LangChain.js