MarkUDown

Agent API — MarkUDown AI 网页浏览智能体 API 文档

POST/api/agent

智能体(Agent)

给智能体一个起始 URL 和一个问题。它会读取页面,询问 LLM 是直接回答还是跟随某个链接,并重复这一过程,直到能够回答或步数用尽。你会得到答案以及它执行的每一步。

参数类型必填默认值描述
urlstring必填—智能体开始的页面。
promptstring必填—交给智能体的问题或任务。
max_stepsinteger可选10智能体最多做出的决策次数(上限 25)。
max_pagesinteger可选5智能体最多访问的页面数(至少 1,上限 15)。任务开始时扣除 max_pages × 2 个积分。
allow_navigationboolean可选true允许智能体跟随链接访问其他页面。
timeoutinteger可选120每个页面的加载超时时间(秒)。
callback_urlstring可选null任务完成时接收结果(POST)的 webhook URL。

请求

curl -X POST https://api.scrapetechnology.com/api/agent \
  -H "X-API-KEY: your-api-key" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://quotes.toscrape.com/", "prompt": "Which tags appear on the quotes by Albert Einstein on the first two pages?", "max_pages": 3}'

响应

// POST returns { "success": true, "id": "abc123", "url": "/api/agent/abc123" }
// Poll GET /api/agent/{id}
{
  "success": true,
  "id": "abc123",
  "status": "completed",
  "data": {
    "success": true,
    "data": {
      "url": "https://quotes.toscrape.com/",
      "answer": "Einstein's quotes are tagged: change, deep-thoughts, ...",
      "steps": [
        {
          "step": 1,
          "url": "https://quotes.toscrape.com/",
          "action": "navigate",
          "reasoning": "Page 1 read; page 2 is needed.",
          "result": "Navigating to: https://quotes.toscrape.com/page/2/"
        },
        {
          "step": 2,
          "url": "https://quotes.toscrape.com/page/2/",
          "action": "answer",
          "reasoning": "Both pages read.",
          "result": "Einstein's quotes are tagged: change, deep-thoughts, ..."
        }
      ],
      "pages_visited": [
        "https://quotes.toscrape.com/",
        "https://quotes.toscrape.com/page/2/"
      ],
      "total_steps": 2
    },
    "processing_time_ms": 21400
  }
}

Python — pip install markudown

from markudown import MarkUDown

md = MarkUDown(api_key="your-api-key")
job = md.agent(
    "https://quotes.toscrape.com/",
    "Which tags appear on the quotes by Albert Einstein on the first two pages?",
    max_pages=3,
)
print(job["data"]["data"]["answer"])

Node.js — npm install markudown-sdk

import { MarkUDown } from "markudown-sdk";

const md = new MarkUDown({ apiKey: "your-api-key" });
const job = await md.agent(
  "https://quotes.toscrape.com/",
  "Which tags appear on the quotes by Albert Einstein on the first two pages?",
  { maxPages: 3 },
);
console.log(job.data.data.answer);
智能体会读取页面(滚动以加载延迟内容)并跟随链接;它不会点击按钮或填写表单。每一步的 action 为 answer、navigate、extract 或 done(出错时为 scrape_failed 或 llm_error)。使用 DELETE /api/agent/{id} 取消运行中的任务。

费用: max_pages × 2 个积分,在任务开始时扣除。