# Why can't ChatGPT or Claude read a WeChat article link, and how do I fix it?

> AI assistants fail to open mp.weixin.qq.com links. Why WeChat blocks them, and how to give Claude or another MCP client a tool that reads WeChat articles.

Canonical page: https://okdata.online/answers/ai-read-wechat-article-link/

**Short answer: **WeChat answers automated fetchers from data-center IPs with a verification page, so the assistant sees no article. Give it a tool that can: connect the okdata MCP server and the assistant can call Official Account article detail, $0.02 per article, to get the title and full text. Ten links a day for a month cost $6.00.

You paste a WeChat article link into ChatGPT or Claude, ask for a summary, and get an apology: the page couldn't be loaded, or the assistant summarises a verification screen. The assistant isn't broken. Its browsing tool fetches pages from data-center servers, and WeChat answers those requests with a page asking to verify the environment, or with HTML whose article body only appears after scripts run. The fix is to give the assistant a tool that returns the article itself: [Official Account article detail](/apis/wechat_mp/articles/get/) takes the link and returns the title, the full text, the account and the publish time.

## Connect it to Claude Code

Add the okdata MCP server to `.mcp.json` in your project (or `~/.claude.json` for every project), with the key in the `OKDATA_API_KEY` environment variable:

```json
{
  "mcpServers": {
    "okdata": {
      "type": "http",
      "url": "https://api.okdata.online/mcp",
      "headers": { "Authorization": "Bearer ${OKDATA_API_KEY}" }
    }
  }
}
```

Restart, run `/mcp`, and the server should show as connected. Other clients that accept a URL plus a header, such as Cursor, connect the same way; the [MCP page](/mcp/) explains how agents use the tools.

## Then just ask

- "Summarise this article in five bullet points: https://mp.weixin.qq.com/s/…"
- "Compare these three articles and tell me where they disagree."
- "What has the account 36氪 published this week?" The assistant finds the account with [Official Account search](/apis/wechat_mp/accounts/search/), then lists its newest articles with [article history](/apis/wechat_mp/accounts/articles/list/).

The assistant uses the free `find` and `describe` tools to pick the endpoint, then `call` to fetch. Only `call` costs money.

## Cost

| Use | Calls | Cost |
|---|---|---|
| One article | 1 | $0.02 |
| 10 links a day for 30 days | 300 | $6.00 |
| An account's week: search, one page of history, 5 articles | 1 + 1 + 5 | $0.72 (search at most) |

## Gotchas

- **Text only.** The assistant gets the words and the cover image link, not the charts inside the article. If a chart carries the point, look at it yourself.
- **Prompt injection.** Articles are written by strangers. Tell the assistant to treat article text as material to analyse, never as instructions.
- **Deleted articles.** They return `not_found` and cost nothing; the assistant should say so instead of guessing.
- **Ask for a budget.** For long research tasks, have the assistant state how many calls it plans before it starts.

## Simpler options

For one article, copy the text in the WeChat app and paste it into the chat. It's free and takes ten seconds. Browser-based assistants that drive your own logged-in browser can also read WeChat pages you open. The MCP route is worth it when you hand the assistant links all day, or when you want it to fetch articles on its own as part of a [content research](/solutions/content-research/) job.

## Endpoints used

- [Official Account article detail](https://okdata.online/apis/wechat_mp/articles/get/) `wechat_mp.articles.get`: $0.02 per call
- [Official Account article history](https://okdata.online/apis/wechat_mp/accounts/articles/list/) `wechat_mp.accounts.articles.list`: $0.04 per call · about 10 articles per page
- [Official Account search](https://okdata.online/apis/wechat_mp/accounts/search/) `wechat_mp.accounts.search`: $0.058 per account · at most $0.58 per call

Updated 2026-11-12