Crawl your website from ChatGPT or Claude

Name your site in the chat

Ask the assistant to crawl your site. If the site is not in your My Sites list yet, the assistant adds it first, within your plan's site limit. You can also add it in the dashboard before you ask.

PerfBee usually starts a first crawl of a new site by itself. The first Detailed crawl of your first site, up to 250 pages, costs no credits. It is given once per account and site, after you confirm your email.

The connector works on the Free plan. When you connect, you choose whether the assistant may start crawls or only read. It can add a site, but it cannot remove sites, change schedules or delete anything.

The steps to connect Claude and ChatGPT are on the MCP page.

The tools crawl, read and check your fixes

You ask in plain words and the assistant picks the tool. Here is what each one does.

• list_sites lists your sites, with the latest crawl of each and the credits left this month.
• start_crawl starts a crawl of a site and returns at once. A site that is not in My Sites yet is added first.
• get_scan_status follows a crawl: its progress while it runs and its result when it ends.
• get_scan_issues reads the broken links, SEO issues and AI readiness issues of a crawl. A broken link comes with the page and the HTML element it sits in.
• get_ai_readiness reads which AI crawlers the robots.txt of a site lets in, and the page issues that can keep pages out of AI answers.
• get_fixes reads what to fix first: each issue grouped by page template, with why it ranks there and how to fix it.
• get_fix_pages reads where one fix goes: the pages, the HTML element to change and what each page has there today.
• check_fix reads the pages of one fix again after you change the code, to see whether it is fixed. It spends credits.
• find_pages searches the page list of your latest crawls by status, issue, folder or text in the URL.
• recheck_url checks one URL of your site now: its status, redirects and main tags, and whether it still links to a URL.
• compare_crawls reads what changed since the previous crawl: new and fixed broken links, and after a detailed crawl, new, gone and changed pages.

Prompts to start with

Replace your-site.com with your site. Each prompt shows the tools it calls.

1. List my sites.

   Calls this tool: list_sites

2. Crawl your-site.com with up to 200 pages.

   Calls this tool: start_crawl

3. Is the crawl of your-site.com done?

   Calls this tool: get_scan_status

4. Which links are broken on your-site.com, and which element do I change?

   Calls this tool: get_scan_issues

5. What should I fix first on your-site.com? Group it by template.

   Calls this tool: get_fixes

6. Which pages of your-site.com miss a meta description, and what do they have now?

   Calls these tools: get_fixes get_fix_pages

7. I fixed the missing alt text on your-site.com. Check if the fix worked.

   Calls these tools: get_fixes check_fix

8. Which pages under /blog/ on your-site.com cannot be indexed?

   Calls this tool: find_pages

9. Check your-site.com/pricing again now. Does it still link to /old-page?

   Calls this tool: recheck_url

10. What changed since the previous crawl of your-site.com?

    Calls this tool: compare_crawls

11. Which AI crawlers does the robots.txt of your-site.com let in?

    Calls this tool: get_ai_readiness

A crawl runs in the background for a few minutes. Ask about it later instead of waiting in the chat. Most prompts read a finished crawl, so crawl the site first if it has none. The compare_crawls tool needs two finished crawls of the same site in the same mode. The get_ai_readiness tool reads the report of a Detailed crawl. That is what start_crawl runs unless you ask for a fast one.

Crawls spend credits and reading is free

A crawl you start from the chat is an ordinary PerfBee crawl. It spends credits at the same rates as a crawl you start in the dashboard, for the pages it checks. check_fix also reads pages again and spends credits the same way. recheck_url checks one URL and costs nothing. The other tools only read, and reading costs no credits.

• A crawl checks at most 500 pages, and 100 unless the assistant asks for another size.
• A crawl needs enough credits for all the pages it may check before it starts.
• Pages a site refuses are listed as Not checked and are not charged.

The other limits, such as crawls a day and calls a minute, are on the MCP page.

Some results stay out of the chat

Numbers that come from Google and accessibility results are never returned by these tools. You see them in your own dashboard and downloads.

What the tools return goes to the AI app and its provider, under your own agreement with them. PerfBee does not send it anywhere else. Your data and permissions lists the rest.

Other ways in

Claude Code and Cursor can connect with an API key; the steps are on the MCP page. To call PerfBee from your own code, use the REST API and signed webhooks. Both are open to every plan, and crawls started either way spend credits.