How to export Xiaohongshu comments for a keyword

The job: you are about to launch a product, or you sell against one, and you want to know what Xiaohongshu users actually say about the category. Notes show what creators want you to think; the comments under them show what readers ask, doubt and compare. So you search a keyword, keep the notes that drew the most discussion, and read their comment sections a page at a time. You end with one CSV: note title, comment text, commenter, likes, reply count, time and the IP region Xiaohongshu shows next to each comment.

No Xiaohongshu account, cookie or xsec_token is involved. Each call is an HTTPS request with your key.

Step by step

Step 1: find the notes worth reading

Call Xiaohongshu note search with query, written the way people type it into the app: 防晒霜, 防晒霜 油皮, 敏感肌 防晒. Two parameters matter here:

• sort: general (the default, Xiaohongshu's own mix), latest or popular. For comment mining, popular surfaces the notes with the most engagement, which are usually the ones with real discussion.
• type: all (the default), video or image. Leave it at all unless you only care about one format.

One call returns about 20 notes for $0.02. Each has id (24 hex characters), title, the start of the text, author, published_at and metrics with likes, favorites (saves) and comments. Sort on metrics.comments to pick the notes to open.

curl -G "https://api.okdata.online/v1/xiaohongshu/notes/search" \
  -H "Authorization: Bearer $OKDATA_API_KEY" \
  --data-urlencode 'query=防晒霜' \
  --data-urlencode 'sort=popular' \
  --data-urlencode 'fields=id,title,metrics.comments,metrics.likes,published_at'

fields keeps the response short without changing the price.

Step 2: page through the comments

Pass a note's id to Xiaohongshu note comments. One call returns about 10 top-level comments for $0.02. Each has text, author (name and ID), likes, replies (how many replies sit under it), published_at and ip_location. parent_id is null because these are top-level comments. For the next page, send page.next_cursor back as cursor, unchanged; stop when page.has_more is false.

curl -G "https://api.okdata.online/v1/xiaohongshu/notes/comments/list" \
  -H "Authorization: Bearer $OKDATA_API_KEY" \
  --data-urlencode 'id=672a4f1d000000001b03c58e'

Comment pages are small on Xiaohongshu, so budget by pages, not by notes: 1,000 comments take about 100 calls.

Step 3: put it together in Python

The script searches once, keeps the 3 notes with the most comments, reads up to MAX_PAGES pages under each and writes one CSV.

import csv
import os
import requests

API = "https://api.okdata.online"
HEADERS = {"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"}
NOTES = 3      # notes to open
MAX_PAGES = 2  # comment pages per note; each page is one call


def get(path, **params):
    body = requests.get(API + path, headers=HEADERS, params=params, timeout=60).json()
    if body["code"] != "ok":
        raise RuntimeError(f"{body['code']}: {body['message']} (charged: {body['charged']})")
    return body


notes = get("/v1/xiaohongshu/notes/search", query="防晒霜", sort="popular")["data"]["items"]
notes = sorted(notes, key=lambda n: n["metrics"]["comments"] or 0, reverse=True)[:NOTES]

with open("xhs_comments.csv", "w", newline="", encoding="utf-8-sig") as f:
    out = csv.writer(f)
    out.writerow(["note_id", "note_title", "comment", "user", "likes", "replies", "ip_location", "published_at"])
    for n in notes:
        cursor = None
        for _ in range(MAX_PAGES):
            params = {"id": n["id"], **({"cursor": cursor} if cursor else {})}
            body = get("/v1/xiaohongshu/notes/comments/list", **params)
            for c in body["data"]["items"]:
                out.writerow([n["id"], n["title"], c["text"], c["author"]["name"], c["likes"],
                              c["replies"], c["ip_location"], c["published_at"]])
            if not body["page"]["has_more"]:
                break
            cursor = body["page"]["next_cursor"]

Raise NOTES and MAX_PAGES to go wider and deeper. Ten notes at five pages each is 50 comment calls, $1.00 on top of the search. Per 1,000 comments the bill comes to about $2.00.

Good to know

• Counts will not match the note. The comment count on a note includes replies. This endpoint lists top-level comments and gives each one's reply count in replies, not the reply text, so you collect fewer rows than metrics.comments.
• Order. Comments come in Xiaohongshu's default order; there is no sort parameter, and no newest-first option.
• IP region. ip_location is the province or country label Xiaohongshu prints next to a comment. It is null when Xiaohongshu shows none.
• No token juggling. You pass the note ID only. Why that matters on Xiaohongshu is in the xsec_token answer.
• Deleted notes are free. A note that is gone or private returns not_found and costs nothing; skip it.
• Retries without paying twice. Send an Idempotency-Key header with each call. The same request with the same key within 24 hours returns the stored answer with meta.replayed: true and no new charge.

To export every comment under one note you already have, see how to export a Xiaohongshu note's comments. For turning comments into a list of complaints and questions, see finding pain points in Douyin and Xiaohongshu comments with an LLM.

{
  "endpoint": "xiaohongshu.notes.search",
  "params": {
    "query": "防晒霜 推荐"
  },
  "idempotency_key": "<a new UUID>"
}