Moonlit

Quick Read · AI Assistants

Ask Claude About Your Website: Using AI Assistants With Your Own Site Data

AI assistants can now read data from the tools you already use. Here's how that works, what to ask, and what to look for before you connect anything.

By the Moonlit teamPublished 2 min read

Short answer

Through connectors built on the Model Context Protocol, assistants like Claude and ChatGPT can read data from tools you've approved, such as your website results. You ask in plain language, and the assistant answers from your saved data with the evidence attached. Before connecting, check what the connector can read, whether it's read-only and how to disconnect it.

Key Takeaways

  • Connectors let an AI assistant read data from tools you've approved.
  • Many are built on the Model Context Protocol, an open standard.
  • Ask specific questions, and ask for the evidence behind each answer.
  • Before connecting, check what it can read, whether it's read-only and how to disconnect.

Asking a chatbot "what should I fix on my website?" used to get you a generic checklist. Fine, but not about your site.

That's changing. AI assistants can now connect to tools you already use and answer from your real data.

How It Works

Assistants like Claude and ChatGPT support connectors: approved links to other apps. Many are built on the Model Context Protocol (MCP), an open standard for connecting AI applications to outside tools and data.

You add a connector, sign in to the other tool and approve what the assistant can see. After that, you can ask questions in plain language and the assistant looks things up in that tool to answer.

Illustrative chat with an AI assistant connected to website results A mock chat. The person asks what to fix first on their website. The assistant answers using saved results: it names the website and the scan date, says the top Quick Win is adding missing page descriptions to four pages, lists the supporting finding, and notes that analytics is not connected so traffic impact is unknown. A footer says the connection is read-only. ILLUSTRATIVE EXAMPLE What should I fix first on my site? Based on your example.com scan from Oct 20: TOP QUICK WIN Add page descriptions to 4 service pages Evidence: 4 pages have no meta description · found on the latest scan Heads up: analytics isn't connected, so I can't tell which page gets the most traffic yet. 🔒 Read-only: the assistant can read saved results, not change anything.
An illustrative example. A good answer names the site and scan date, shows its evidence and admits what it doesn't know.

What to Ask

Good questions are specific and ask for evidence:

Asking "show me the evidence" is a great habit. It keeps the answer tied to your actual data instead of general advice.

What to Check Before You Connect

Be cautious if

  • It can change or delete things
  • It's unclear what data it reads
  • There's no obvious way to disconnect

Look for

  • Read-only access where possible
  • A clear list of what it can see
  • A one-click disconnect

Connectors work on your behalf with your sign-in, so treat them like any app you'd give account access to.

Using Moonlit With Claude and ChatGPT

Moonlit has a read-only connector for Claude and ChatGPT. Once connected, the assistant can read your saved results, like your Website Health score, recommendations, the findings behind them and your Quick Wins. It can't start scans, change tasks or edit your account. Free plans include your score and up to three Quick Wins with their evidence.

Setup steps and the full list of what it can read are on our Claude and ChatGPT page.

Frequently Asked Questions

Is it safe to connect my data?

It depends on the connector. Prefer read-only access, check what's included and know how to disconnect. Moonlit's connector is read-only and can be disconnected in Settings.

Will the assistant make things up?

Assistants can still make mistakes. Asking for the evidence behind each answer, and checking it, is the best protection.

Sources