公开 API 文档

curl -X POST "https://api.dk-ai.tech/api/tools/dy_search_videos" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-binary '{"keyword":"\u5496\u5561","limit":20}'

{
  "params": {
    "keyword": "咖啡",
    "limit": 20
  }
}

{
  "mcpServers": {
    "dk-social": {
      "url": "https://api.dk-ai.tech/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

{
  "items": [],
  "resultCount": 20,
  "pagination": {
    "hasMore": true,
    "nextCursor": "20",
    "cursorParam": "cursor",
    "searchId": "202607...",
    "requiredNextParams": ["cursor", "search_id"],
    "pageSize": 20
  },
  "nextCursor": "20"
}

# 抓 5 页
for page in range(1, 6):
    payload = {"keyword": "咖啡", "limit": 20}
    if cursor:
        payload["cursor"] = cursor
    if search_id:
        payload["search_id"] = search_id
    data = call_dk_tool("dy_search_videos", payload)
    append_items_to_csv(data["items"])
    cursor = data.get("pagination", {}).get("nextCursor")
    search_id = data.get("pagination", {}).get("searchId") or search_id
    if not cursor or not data.get("pagination", {}).get("hasMore"):
        break

20 页任务：拆成 4 批，每批 5 页；每页写 CSV/JSONL；每批写 checkpoint。
50 页任务：拆成 10 批，每批 5 页；智能体只读取阶段汇总和最终文件，不要把原始 JSON 全部塞进上下文。
checkpoint 字段：toolName, page, cursor, nextCursor, searchId, traceId, creditsUsed, resultCount。