Mastra

npm install @mastra/core firecrawl zod

FIRECRAWL_API_KEY=your_firecrawl_key
OPENAI_API_KEY=your_openai_key

import { createWorkflow, createStep } from "@mastra/core/workflows";
import { z } from "zod";
import { Firecrawl } from "firecrawl";
import { Agent } from "@mastra/core/agent";

const firecrawl = new Firecrawl({
  apiKey: process.env.FIRECRAWL_API_KEY || "fc-YOUR_API_KEY"
});

const agent = new Agent({
  name: "summarizer",
  instructions: "You are a helpful assistant that creates concise summaries of documentation.",
  model: "openai/gpt-5-nano",
});

// 步骤 1:使用 Firecrawl SDK 搜索
const searchStep = createStep({
  id: "search",
  inputSchema: z.object({
    query: z.string(),
  }),
  outputSchema: z.object({
    url: z.string(),
    title: z.string(),
  }),
  execute: async ({ inputData }: { inputData: { query: string } }) => {
    console.log(`搜索中:${inputData.query}`);
    const searchResults = await firecrawl.search(inputData.query, { limit: 1 });
    const webResults = (searchResults as any)?.web;

    if (!webResults || !Array.isArray(webResults) || webResults.length === 0) {
      throw new Error("未找到搜索结果");
    }

    const firstResult = webResults[0];
    console.log(`找到:${firstResult.title}`);
    return {
      url: firstResult.url,
      title: firstResult.title,
    };
  },
});

// 步骤 2:使用 Firecrawl SDK 抓取 URL
const scrapeStep = createStep({
  id: "scrape",
  inputSchema: z.object({
    url: z.string(),
    title: z.string(),
  }),
  outputSchema: z.object({
    markdown: z.string(),
    title: z.string(),
  }),
  execute: async ({ inputData }: { inputData: { url: string; title: string } }) => {
    console.log(`抓取中:${inputData.url}`);
    const scrapeResult = await firecrawl.scrape(inputData.url, {
      formats: ["markdown"],
    });

    console.log(`已抓取:${scrapeResult.markdown?.length || 0} 个字符`);
    return {
      markdown: scrapeResult.markdown || "",
      title: inputData.title,
    };
  },
});

// 步骤 3:使用 Claude 生成摘要
const summarizeStep = createStep({
  id: "summarize",
  inputSchema: z.object({
    markdown: z.string(),
    title: z.string(),
  }),
  outputSchema: z.object({
    summary: z.string(),
  }),
  execute: async ({ inputData }: { inputData: { markdown: string; title: string } }) => {
    console.log(`生成摘要中:${inputData.title}`);

    const prompt = `用 2-3 句话总结以下文档:\n\n标题:${inputData.title}\n\n${inputData.markdown}`;
    const result = await agent.generate(prompt);

    console.log(`摘要已生成`);
    return { summary: result.text };
  },
});

// 创建工作流
export const workflow = createWorkflow({
  id: "firecrawl-workflow",
  inputSchema: z.object({
    query: z.string(),
  }),
  outputSchema: z.object({
    summary: z.string(),
  }),
  steps: [searchStep, scrapeStep, summarizeStep],
})
  .then(searchStep)
  .then(scrapeStep)
  .then(summarizeStep)
  .commit();

async function testWorkflow() {
  const run = await workflow.createRunAsync();
  const result = await run.start({
    inputData: { query: "Firecrawl documentation" }
  });

  if (result.status === "success") {
    const { summarize } = result.steps;

    if (summarize.status === "success") {
      console.log(`\n${summarize.output.summary}`);
    }
  } else {
    console.error("工作流失败:", result.status);
  }
}

testWorkflow().catch(console.error);