Pular para o conteúdo principal
Este guia mostra como integrar o Firecrawl ao LangGraph para criar fluxos de trabalho de agentes de IA capazes de extrair e processar conteúdo da web.

​
Configuração

npm install @langchain/langgraph @langchain/openai @mendable/firecrawl-js
Crie o arquivo .env: 
FIRECRAWL_API_KEY=sua_chave_firecrawl
OPENAI_API_KEY=sua_chave_openai
Observação: Se estiver usando Node < 20, instale o dotenv e adicione import 'dotenv/config' ao seu código.

​
Fluxo básico

Este exemplo demonstra um fluxo básico do LangGraph que captura dados de um site e analisa o conteúdo. 
import FirecrawlApp from '@mendable/firecrawl-js';
import { ChatOpenAI } from '@langchain/openai';
import { StateGraph, MessagesAnnotation, START, END } from '@langchain/langgraph';

// Inicializar o Firecrawl
const firecrawl = new FirecrawlApp({ apiKey: process.env.FIRECRAWL_API_KEY });

// Inicializar o LLM
const llm = new ChatOpenAI({
    model: "gpt-5-nano",
    apiKey: process.env.OPENAI_API_KEY
});

// Definir o nó de scrape
async function scrapeNode(state: typeof MessagesAnnotation.State) {
    console.log('Fazendo scrape...');
    const result = await firecrawl.scrape('https://firecrawl.dev', { formats: ['markdown'] });
    return {
        messages: [{
            role: "system",
            content: `Conteúdo extraído: ${result.markdown}`
        }]
    };
}

// Definir o nó de análise
async function analyzeNode(state: typeof MessagesAnnotation.State) {
    console.log('Analisando...');
    const response = await llm.invoke(state.messages);
    return { messages: [response] };
}

// Construir o grafo
const graph = new StateGraph(MessagesAnnotation)
    .addNode("scrape", scrapeNode)
    .addNode("analyze", analyzeNode)
    .addEdge(START, "scrape")
    .addEdge("scrape", "analyze")
    .addEdge("analyze", END);

// Compilar o grafo
const app = graph.compile();

// Executar o workflow
const result = await app.invoke({
    messages: [{ role: "user", content: "Resumir o site" }]
});

console.log(JSON.stringify(result, null, 2));

​
Fluxo em várias etapas

Este exemplo demonstra um fluxo de trabalho mais complexo que faz a raspagem de várias URLs e as processa. 
import FirecrawlApp from '@mendable/firecrawl-js';
import { ChatOpenAI } from '@langchain/openai';
import { StateGraph, Annotation, START, END } from '@langchain/langgraph';

const firecrawl = new FirecrawlApp({ apiKey: process.env.FIRECRAWL_API_KEY });
const llm = new ChatOpenAI({ model: "gpt-5-nano", apiKey: process.env.OPENAI_API_KEY });

// Definir estado personalizado
const WorkflowState = Annotation.Root({
    urls: Annotation<string[]>(),
    scrapedData: Annotation<Array<{ url: string; content: string }>>(),
    summary: Annotation<string>()
});

// Fazer scrape de múltiplas URLs
async function scrapeMultiple(state: typeof WorkflowState.State) {
    const scrapedData = [];
    for (const url of state.urls) {
        const result = await firecrawl.scrape(url, { formats: ['markdown'] });
        scrapedData.push({ url, content: result.markdown || '' });
    }
    return { scrapedData };
}

// Resumir todo o conteúdo coletado
async function summarizeAll(state: typeof WorkflowState.State) {
    const combinedContent = state.scrapedData
        .map(item => `Conteúdo de ${item.url}:\n${item.content}`)
        .join('\n\n');

    const response = await llm.invoke([
        { role: "user", content: `Resuma estes sites:\n${combinedContent}` }
    ]);

    return { summary: response.content as string };
}

// Construir o grafo do workflow
const workflow = new StateGraph(WorkflowState)
    .addNode("scrape", scrapeMultiple)
    .addNode("summarize", summarizeAll)
    .addEdge(START, "scrape")
    .addEdge("scrape", "summarize")
    .addEdge("summarize", END);

const app = workflow.compile();

// Executar workflow
const result = await app.invoke({
    urls: ["https://firecrawl.dev", "https://firecrawl.dev/pricing"]
});

console.log(result.summary);
Para mais exemplos, confira a documentação do LangGraph.