wechat-article-crawler

A Claude Code skill for crawling WeChat public account (微信公众号) articles into structured data and clean markdown — with images downloaded locally to bypass hotlink protection.

Features

• Spoofs WeChat in-app browser User-Agent to get full article content
• Fixes lazy-loaded images (data-src → src)
• Extracts structured metadata: title, author, publish time, account description
• Generates clean markdown with inline images (SVG placeholders removed)
• Downloads images locally with correct Referer header to bypass CDN hotlink protection
• Concurrent downloads with rate limiting (semaphore)

Quick Start

As a Claude Code Skill

Copy this directory to .claude/skills/crawl-wechat in your project:

# Clone into your project's skills directory
git clone https://github.com/gxcsoccer/wechat-article-crawler.git .claude/skills/crawl-wechat

Then in Claude Code, just paste a WeChat article URL or say "抓取这篇微信文章".

As a Standalone Script

# Install dependencies
pip install crawl4ai aiohttp && crawl4ai-setup

# Crawl an article with images
python scripts/crawl_wechat.py "https://mp.weixin.qq.com/s/xxx" \
  --download-images \
  --save-markdown \
  --save-html \
  --output-dir ./output

As a Python Library

import asyncio
from scripts.crawl_wechat import crawl_wechat_article

article = asyncio.run(crawl_wechat_article(
    "https://mp.weixin.qq.com/s/xxx",
    images_dir="./output/images",
))

print(article["title"])
print(article["markdown"])  # images reference local paths

Output

CLI Options

usage: crawl_wechat.py [-h] [--save-html] [--save-markdown] [--download-images] [--output-dir DIR] url

positional arguments:
  url                WeChat article URL (mp.weixin.qq.com/s/...)

options:
  --save-html        Save HTML to file
  --save-markdown    Save markdown to file
  --download-images  Download images locally to bypass hotlink protection
  --output-dir DIR   Directory for saved files (default: .)

How It Works

1. User-Agent spoofing — Sets MicroMessenger/8.0.43 so WeChat serves the full article
2. Dynamic wait — wait_for="css:#js_content" ensures the body is rendered
3. Lazy-image JS injection — Copies data-src → src on all <img> tags before scraping
4. Structured extraction — JsonCssExtractionStrategy targets WeChat's DOM (#activity-name, #js_name, #publish_time, #js_content)
5. Markdown cleanup — Removes 1x1 SVG placeholders and URL-encoded data-URI fragments
6. Image download — Fetches from mmbiz.qpic.cn with Referer: https://mp.weixin.qq.com/, replaces URLs in output

Limitations

• Requires a valid, non-expired WeChat article URL — cannot search or list articles
• High-frequency crawling may trigger anti-bot measures (CAPTCHAs, IP blocks)
• Some temporary share links expire after a period

License

MIT