How to Get Xiaohongshu (RedNote) Data Through an API

Get the top Xiaohongshu notes for a keyword as a table, with likes, saves, comments and post dates. Paste this into your agent:

Set up https://monid.ai/SKILL.md, and then use Monid to search Xiaohongshu for the keyword 抹茶拿铁 (matcha latte), most liked first, with tikhub /api/v1/xiaohongshu/app_v2/search_notes. Give me a table of the 20 notes with title, note type, likes, saves, comments, shares and post date, sorted by likes, and tell me how many were posted in the last 30 days. Leave author names out.

One prompt returns the page as rows. 1 min setup · 5-10 hrs saved.

If you run marketing, brand or creator research for a product that sells in China, Xiaohongshu (RedNote outside China) is where buyers compare notes before they buy. The app is in Chinese, search results arrive 20 notes at a time, the sort is the one the app picked, and there is no export button. A weekly check on three keywords is 60 notes copied into a sheet by hand, and the like counts are already stale by the time you finish.

The usual shortcuts do not hold up. A script you write yourself has to reproduce an app that signs its requests and rotates its tokens. A monitoring tool shows a dashboard and keeps the rows. Searching in English misses the notes written the way Chinese users write.

This guide ends with a table of 20 notes per page for any keyword, with likes, saves, comments, shares, post date and note type, from one call. It runs through Monid, the OpenRouter for agent tools: one key, one balance, every tool priced per call. EveryInfra, whose Xiaohongshu data API is one of the four routes below, is a content partner on this guide; the measurements are ours. Here is what the two runs returned on October 29, 2026.

Why Xiaohongshu matters before you plan a China launch

A Xiaohongshu note is a buyer's review with pictures, and the saves count tells you which notes people kept to act on later. Likes measure attention; saves measure intent. The notes that rank for your category are the ones a shopper reads in the store aisle, and they decide which claims in your own content will land and which will be laughed at.

The numbers hide inside the app. You can see one note's likes, saves and comments at a time, but not the distribution for a keyword, how old the top notes are, or whether the conversation moved last month. Those are table questions, so the deliverable here is rows: rerun the same keyword every week, compare your brand term with a competitor's, and see whether the notes driving the category are three weeks old or three years old.

The 4 ways to get Xiaohongshu notes

There are four routes to the rows. Two of them are a maintenance job, and two of them are a call.

Read the app by hand

Open the app, search, scroll, and type the counts into a sheet. This works for one keyword once. Next week the counts have moved, the order has changed, and you cannot tell which notes are new.

Scrape it yourself

A script with a browser or a reverse-engineered client can read the search page. The cost is the maintenance: the app signs its requests, note links carry a token that expires, and the markup changes. You pay that cost again at every change.

A dedicated data API

EveryInfra sells a pay-per-call data API that covers Xiaohongshu alongside 90 other platforms, with one request shape for all of them. On October 29, 2026 its public catalog listed 11 Xiaohongshu actions, search, profile, user posts, note, video, comments, batch comments, comment replies, user search, hashtag and trending, each at $5.56 per 1,000 calls, Verified 2026-10-29 on that catalog. Its Xiaohongshu (RedNote) data API page lists the fields each action returns and its health over the last 24 hours. This is the right route when Xiaohongshu is one of a fixed set of platforms you read every day and you want one vendor and one bill.

Monid

Xiaohongshu search also sits in the Xiaohongshu tools on monid.ai/tools, 49 endpoints on October 29, 2026, next to Douyin, Weibo and TikTok readers, under one key and one balance, priced per call. Two providers offer a keyword search there, and the agent picks one at run time. The social media scraping API guide walks the same four-route choice for TikTok, Instagram and X.

How to get Xiaohongshu notes with Monid

Four steps: install, pick the keyword and sort, run the search, read the table. Monid supplies one key and one balance; an endpoint is the single operation you pick, here a Xiaohongshu keyword search.

Step 1. Install and add your key

For agents. Get an API key at app.monid.ai, then give your agent the key and this setup line. The agent quickstart explains the workflow.

set up https://monid.ai/SKILL.md

For humans. Install the CLI and save your key locally. The CLI quickstart covers the rest.

npm install -g @monid-ai/cli
monid keys add -k <your-api-key> -l main

Step 2. Pick the keyword and the sort

Search in Chinese: most notes about a matcha latte are tagged 抹茶拿铁, not "matcha latte". Then pick the sort by the question you are asking: most liked for "what defines this category", newest for "what changed this week". TikHub's search takes the sort as sort_type and also accepts a note type filter and a time filter for the last day, week or six months.

keyword      抹茶拿铁
sort_type    popularity_descending   (or time_descending, comment_descending, collect_descending)
page         1

Step 3. Run the search

Send the keyword to TikHub's Xiaohongshu search. One call is one page of 20 notes; set page to 2 and pass back the search_id and search_session_id from the first response for the next page.

monid run -p tikhub -e /api/v1/xiaohongshu/app_v2/search_notes \
  --query '{"keyword":"抹茶拿铁","page":1,"sort_type":"popularity_descending"}' \
  -w -j

On October 29, 2026 this call completed in 3.2 seconds with HTTP 200 and 20 notes. The search card on the Xiaohongshu page lists $0.01 per call, Verified 2026-10-29.

The second provider takes the same keyword as a JSON body and sorts by general, hot or time. We ran it on the same keyword with hot, and the difference between the two sorts is the first finding below.

monid run -p anyapi -e /rednote.search \
  -i '{"query":"抹茶拿铁","sort":"hot"}' \
  -w -j

Step 4. Read and export the table

Retrieve the completed run and keep the raw file next to the table you build from it.

monid runs get -r <run-id> -o xiaohongshu-matcha.json

Every TikHub note carries id, title, desc, type (normal or video), liked_count, collected_count, comments_count, shared_count, timestamp, geo_info, images_list, a video_info_v2 block for video notes, and an xsec_token you need to open the note. The author block holds the nickname, red_id and user id; drop it before the table leaves your machine, the analysis below never needed it. AnyAPI returns the same counts under likeCount, collectCount, commentCount and shareCount, the post time as createdUtc, and a nextCursor for the next page.

How do you read the results?

Read the table as a sort question and a time question, because those are the two places a Xiaohongshu result misleads. Everything below comes from the three runs on October 29, 2026; no author data was used.

"Hot" is not "most liked"

The same keyword on the same day returned notes with a median of 2,132 likes under one provider's "most liked" and 97 likes under the other's "hot". Hot leans recent: a quarter of its notes were under 30 days old, against none of the most-liked set. Neither is wrong. They answer different questions, and the rule that follows is to name the sort in the file name, because a "top 20" with no sort attached is not comparable with next week's.

Most liked is old

The most-liked page for 抹茶拿铁 had a median post age of 345 days, and the oldest note was over three and a half years old. If your task says "this week", sort by time or filter on the post timestamp after the call. The notes that define a category and the notes it is reading this week are two different tables.

The English keyword finds a different set

"matcha latte" on the most-liked sort returned 20 notes with a median of 2,961 likes, half of them video, and a median age of 331 days. The Chinese keyword returned 15 image notes and 5 videos, and the two sets barely overlap, because the English query matches notes that carry the English tag, often from creators posting from abroad. Run both when the audience spans China and the diaspora.

How much does it cost?

TikHub's Xiaohongshu search is $0.01 per call, Verified 2026-10-29 against the Xiaohongshu tools page, and one call is one page of 20 notes. AnyAPI's search is $0.012 per call, Verified 2026-10-29, with one route today. Our three runs cost $0.032 in total.

Two billing rules matter more than the figure. First, both searches bill per call, not per note, so a page with 20 notes and a page with 3 cost the same; ask for the sort and filter you want in the first call. Second, discovery and inspection are free, so an agent can look up the endpoint, its parameters and its price on every task without spending anything. A weekly check on three keywords, two sorts each, is six calls a week.

Can an agent use Monid and a dedicated data server together?

Yes. An MCP client connects to more than one server at a time and keeps one client per server, so an agent can hold Monid for the long tail of tools and a dedicated data server for the calls it repeats every morning.

EveryInfra runs a remote server at https://api.everyinfra.com/mcp with 25 tools on October 29, 2026; the data tools are everyinfra_list_capabilities and everyinfra_call_api, and setup is on the EveryInfra MCP server page. A client that reads an mcpServers file takes one block per server.

{
  "mcpServers": {
    "monid": { "url": "https://mcp.monid.ai/v1" },
    "everyinfra": {
      "url": "https://api.everyinfra.com/mcp",
      "headers": { "Authorization": "Bearer <your EveryInfra API key>" }
    }
  }
}

Give the agent a short rule for which server takes which job. The split that works is repetition: a fixed platform set on a fixed schema every day goes to the dedicated server, and everything else, including the first time you touch a new platform, goes through discover, inspect and run.

Daily jobs on the platforms listed below go to the dedicated data server;
use each platform's own language for the keyword and sort by time when
the task says "this week". Everything else, and any platform not listed,
goes through Monid: discover, inspect, run. Log the charge of every call.

Which endpoint for which job?

If the platform you track is TikTok rather than Douyin, the TikTok scraper guide sorts its profile, video and comment endpoints by job in the same shape.

When should you use EveryInfra instead?

Use EveryInfra directly if Xiaohongshu is one of a fixed set of Chinese platforms you read every day and you want one request shape and one bill for all of them. Its catalog on October 29, 2026 covered Douyin, Kuaishou, Bilibili, Weibo, Zhihu, WeChat Channels, Taobao, JD, 1688 and Dewu beside Xiaohongshu, and the Douyin data API page lists creator search, audience and rate-card actions that the Monid catalog does not carry today. Its pay-per-call pricing groups actions into tiers from $0.56 to $5.56 per 1,000 calls, up to $280 per 1,000 for the heaviest, Verified 2026-10-29 on that page; its docs state that failed or empty results are not charged and that a wrong parameter returns a 422 with the allowed values. You are reading the Monid blog; this guide shows what two Monid endpoints returned, not that either route is cheaper.

Two limits are ours to state. The two Monid searches return 20 notes per call and name their sorts differently, so a comparable weekly series needs the same provider and the same sort every time. And comment threads were not part of today's runs; the endpoints exist on the Xiaohongshu tools page, but this guide only shows what the searches returned.

Conclusion

You now have a repeatable call that turns a Xiaohongshu keyword into 20 notes as rows, with likes, saves, comments, shares, post date and note type, and a reading of the result that keeps "most liked", "hot" and "this week" as three separate tables. Run it on a schedule through Monid, name each file by keyword and sort, and the next launch brief can quote the notes buyers actually saved. The 5 to 10 hours saved is our estimate for writing and then keeping alive a client for an app that signs its requests and expires its note tokens; the setup line is a one-minute target from key to first call. What you keep is a weekly answer to "what is the category reading" without opening the app.

FAQ

Published October 2026.