第 10 期：HTTP 请求节点与外部 API 集成

HTTP 请求节点的威力

HTTP 节点让 Workflow 能够与整个互联网交互：查天气、调支付、发邮件、存数据库。

graph LR
    LLM[LLM 提取参数] --> HTTP[HTTP 请求节点]
    HTTP --> Parse[Code 解析响应]
    Parse --> LLM2[LLM 总结结果]
    
    HTTP -.-> ExtAPI[外部 REST API]

配置详解

GET 请求：查询天气

# 节点配置
Method: GET
URL: https://api.openweathermap.org/data/2.5/weather
Query Parameters:
  q: "{{#start.city#}}"
  appid: "{{#env.WEATHER_API_KEY#}}"
  units: metric
  lang: zh_cn
Headers:
  Accept: application/json
Timeout: 10

POST 请求：调用 Slack Webhook

Method: POST
URL: https://hooks.slack.com/services/T00/B00/xxx
Headers:
  Content-Type: application/json
Body (JSON):
  text: "{{#llm_1.text#}}"
  channel: "#ai-notifications"

带 Bearer Token 鉴权

Method: GET
URL: https://api.github.com/repos/langgenius/dify
Headers:
  Authorization: "Bearer {{#env.GITHUB_TOKEN#}}"
  Accept: application/vnd.github.v3+json

响应解析

# Code 节点：解析 HTTP 响应
import json

def main(http_response: str, status_code: int) -> dict:
    if status_code != 200:
        return {
            "success": False,
            "error": f"API returned {status_code}",
            "data": None
        }
    
    data = json.loads(http_response)
    
    return {
        "success": True,
        "temperature": data["main"]["temp"],
        "description": data["weather"][0]["description"],
        "city": data["name"]
    }

实战：构建一个新闻聚合 Workflow

graph TB
    Start[用户输入关键词] --> HTTP1[HTTP: 新闻API搜索]
    HTTP1 --> Code1[Code: 提取前5条]
    Code1 --> Iterator[Iterator: 遍历每条新闻]
    
    subgraph "Iterator 内部"
        LLM1[LLM: 生成摘要]
        LLM2[LLM: 情感分析]
        LLM1 --> LLM2
    end
    
    Iterator --> Agg[Aggregator: 汇总]
    Agg --> LLM3[LLM: 生成综合报告]
    LLM3 --> End[End: 输出报告]

# 代码节点：提取前 5 条新闻
def main(api_response: str) -> dict:
    import json
    data = json.loads(api_response)
    articles = data.get("articles", [])[:5]
    
    return {
        "articles": [
            {
                "title": a["title"],
                "source": a["source"]["name"],
                "url": a["url"],
                "content": a.get("content", "")[:500]
            }
            for a in articles
        ],
        "count": len(articles)
    }