简体中文 | English

REDnote Clipper

Clip one or more Xiaohongshu posts into editable, plain-text Markdown.

REDnote Clipper (named XHS Clipper in the extension interface) is a Chrome extension for saving Xiaohongshu image posts as editable Markdown. The floating workspace handles the current post or a selection of up to 20 posts from the home feed, search results, and creator profiles. Select the extension button in the Chrome toolbar to open or collapse the workspace. OCR runs locally in Chrome. Each result can be downloaded as a .md file or written to a local Obsidian vault.

Features

• Capture the title, body, author, tags, publish time, engagement counts, source URL, and image order.
• Scan the post cards already loaded on the current page, filter titles by keyword, and select up to 20 posts per batch.
• Run optional local OCR on all images, everything except the cover, or a custom page range.
• Follow per-post progress and remaining time, or stop safely after the current post.
• Repair common Chinese-English spacing, broken lines, paragraph breaks, and exact OCR duplicates.
• Download a .md file or write it to Obsidian through a local REST API.
• Handle an existing Obsidian note by skipping it, saving a new version, or overwriting the matching source note.
• Remember OCR, output, section, and collapsed-window choices in the current Chrome profile.
• Follow Chrome's display language: zh_CN uses Simplified Chinese, zh_TW uses Traditional Chinese, and every other locale uses English.

OCR runs with bundled PP-OCRv6 Tiny models. Images and recognized text are not sent to a remote OCR service.

Performance

Updated: 2026-08-31

Using browser-based OCR and long-image tiling, REDnote Clipper batch-converted REDnote posts and image text into editable Markdown, with a mean image-text extraction time of 684.1 ms per image in a 120-post test.

Performance record.

Installation

Install the packaged extension

1. Open the latest release and download xhs-clipper-v1.0.1.zip from Assets. Do not download GitHub's automatically generated Source code archives.
2. Extract the ZIP to get the xhs-clipper-v1.0.1 folder. Chrome cannot load the ZIP directly.
3. Open chrome://extensions/ and enable Developer mode.
4. Select Load unpacked, then choose the extracted xhs-clipper-v1.0.1 folder.

The packaged extension does not require Node.js or pnpm. After installation, pin XHS Clipper to the Chrome toolbar and reload any Xiaohongshu tabs that were already open.

When upgrading an unpacked copy, copy the new files over the folder Chrome already loads and select Reload on the extension card. This keeps the extension ID and local settings. If you load the new version folder separately, disable or remove the old copy first so two floating workspaces do not appear.

Build from source

Building from source requires Node.js 20 or newer and pnpm.

pnpm install
pnpm verify

The build is written to the dist directory. Load that directory from chrome://extensions/.

Floating clipping workspace

The floating workspace is available on the Xiaohongshu home feed, search results, creator profiles, and individual post pages. Select the XHS Clipper button in the Chrome toolbar to open or collapse it. An individual post is prepared automatically on its article page. On other pages, select Scan current page to read only the cards already loaded in that tab. Scroll first and scan again if you need more results. Existing results are cleared when the page route changes.

Enter one or more keywords separated by spaces. Every keyword must appear in the normalized title. Select individual posts or choose the first 20, review the OCR and output summary, then start the batch. XHS Clipper opens the selected posts one at a time in inactive tabs and closes each tab after processing it.

The panel shows the current stage, elapsed time, estimated time remaining, and the result for each post. The stop button finishes the current post before canceling the rest. Selecting a post title opens that post in the current tab.

The first OCR job initializes the local ONNX runtime. Later jobs reuse the loaded engine. Long images are divided into smaller sections and merged in source order. If one image fails, the extension continues with the remaining images. Post-processing repairs layout without rewriting the source.

Automated access can still trigger platform rate limits or account checks. Keep batches small, avoid repeated runs, and stop if Xiaohongshu asks for login verification or a captcha.

Write to Obsidian

Direct writing uses the Local REST API with MCP community plugin. Obsidian must be open when XHS Clipper writes a note.

1. Get the local API address and key

1. In Obsidian, open Settings > Community plugins > Browse.
2. Install and enable Local REST API with MCP.
3. Open its settings and enable the non-encrypted HTTP server.
4. Copy the Non-encrypted (HTTP) API URL and API key.

View the Obsidian plugin settings example

2. Configure XHS Clipper

1. Select Configure Obsidian permissions and export path in the floating workspace.
2. Confirm the shared export folder, Clippings/XHS.
3. Enter the local API address. The default is http://127.0.0.1:27123.
4. Paste the API key, save the settings, and select Test connection.

The API key stays in extension storage for the current Chrome profile. XHS Clipper only accepts `127.0.0.1` and `localhost` addresses. Do not publish the API key in screenshots, issues, or logs.

Markdown output

Each file contains YAML frontmatter, the written post, and OCR text in image order. The default filename is title_author_date.md. XHS Clipper exports text only. It does not download or embed the original images.

---
title: "Example note"
source: "https://www.xiaohongshu.com/explore/example"
platform: "xiaohongshu"
author: "Example author"
ocr: true
---

# Example note

## Post

Post text...

## Image text

### Image 1

Recognized text...

Responsible use

XHS Clipper is an independent, unofficial project. It is not affiliated with, endorsed by, or sponsored by Xiaohongshu. Xiaohongshu and related names and marks belong to their respective owners.

Use the extension only for content you are permitted to access and process. You are responsible for following applicable laws, copyright rules, and platform terms. Do not republish, sell, or distribute other people's content without permission.

Development

pnpm test
pnpm build
pnpm verify
pnpm benchmark:job
pnpm benchmark:ocr

The OCR benchmark is built and served separately so benchmark assets never enter the extension package. Set CHROME_PATH if Chrome is not installed at the platform default path. Use OCR_BENCHMARK_ITERATIONS and OCR_BENCHMARK_WARMUP to change the run length.

The code is split into page extraction, OCR, text post-processing, Markdown export, and Obsidian writing. Future search or Skill entry points can reuse the existing Note data structure.

OCR uses Apache-2.0 licensed PP-OCRv6 Tiny models and the MIT licensed onnxruntime-web runtime. Image preprocessing and result decoding follow the public behavior of PaddleOCR.js, with Canvas and TypeScript replacing OpenCV. Product and data-structure references include Obsidian Web Clipper, Bilibili Obsidian Clipper, ChatGPT Exporter, xiaohongshu-mcp, and xiaohongshu-skills. Their application code was not copied into this project.

XHS Clipper is licensed under the Apache License 2.0. See THIRD_PARTY_NOTICES.md for model, runtime, and attribution details. Version history is in CHANGELOG.md.