How to Scrape a Website in ChatGPT
ChatGPT browsing can read a page, but a read/fetch ToolRouter MCP app gives you a repeatable structured scrape from a URL. This path works on ChatGPT web: Pro supports read/fetch MCP apps, while Business and Enterprise/Edu support full MCP. Web Scraper is the primary catalog tool for this job.
The Claude version of this workflow lives at How to Scrape Websites with Claude. Same catalog, different client setup.
Access only content you are permitted to collect. Respect site terms, robots directives, rate limits, privacy, and applicable law.
OpenAI's current guide does not list Free or Plus for custom MCP apps, and MCP apps are not available on mobile. Source: OpenAI's current MCP availability guide.
What ChatGPT Can and Cannot Do Natively for Scraping
| What you want | ChatGPT alone | ChatGPT + ToolRouter |
|---|---|---|
| Browse or summarize a public page | Often yes (browsing / search modes) | Yes, with an explicit fetch |
| Crawl multiple pages with a declared limit | Not a general crawler contract | Yes, within tool limits |
| Extract structured fields (name, price, SKU…) | Ad hoc from the summary | Yes, with a schema or plain-English prompt |
| Handle JavaScript-heavy public pages | Varies | Use rendering when the tool supports it |
| Export a clean markdown or typed JSON result | Unreliable as a pipeline | Yes |
How to Set Up Scraping in ChatGPT
On Business, an admin or owner must create and publish the app. On Enterprise/Edu, an admin grants Developer mode access and an authorized developer can test it. Pro users can create a read/fetch app for their own web account.
- Go to Settings → Apps → Advanced settings and enable Developer mode.
- Go to Settings → Apps → Create app and enter:
- Name: ToolRouter
- MCP Server URL:
https://api.toolrouter.com/mcp - Click Scan Tools, review the returned tools, then create the app.
- In a web chat, select ToolRouter from the tools menu for the message, then ask for the scrape.
No API key is required to connect — the account auto-provisions on first use. Full steps also live in Connect ChatGPT and Can ChatGPT use MCP tools?.
Connect your assistant, then try the task described in this guide.
The Prompt to Type
Scrape https://example.com/pricing and return a structured list of plan names,
prices, and included features as JSON.Or for a bounded crawl:
Crawl https://example.com/blog (max 20 pages) and give me each post title,
URL, and publish date in a table.ChatGPT should call Web Scraper. For harder public pages, Stealth Scraper may be offered as an alternative rendering path — it does not override site rules or guarantee access.
When We Are Not the Answer
If you only need a one-off summary of a public article, ChatGPT browsing may be enough. If you need a production crawler with your own proxies, queues, and legal review, write or run a dedicated scraping stack. ToolRouter helps when you want structured extraction *inside* ChatGPT without leaving the chat.
Frequently Asked Questions
Can ChatGPT scrape websites natively?
**Browsing ≠ structured scraping.** ChatGPT can often read and summarize a page. It does not give you a reliable crawl-and-schema pipeline on its own.
Do I need Claude Code?
**No.** This is a ChatGPT path. Use Developer mode → Create app. Claude Code is unrelated.
Do I need an API key?
**No.** ToolRouter auto-provisions on first use after you create the MCP app.
Is this the same as the Claude scrape guide?
Same idea, different setup. See [How to Scrape Websites with Claude](/blog/how-to-scrape-websites-with-claude) for the Claude connector version. Connect [ToolRouter for ChatGPT](/connect?client=chatgpt), select the app for your message, and ask for structured fields from the URL.
