How do I tell the AI which Safari tab to read?

Identify the tab by a partial URL with url_match (for example "github.com") or by its numeric index. The AI passes that to safari_read_tab and reads the matching tab.

Does safari_read_tab change anything on the page?

No. It is a read-only tool. It only extracts the rendered text content of the tab and never clicks, types, scrolls, or modifies the page.

Can it read pages behind a login?

Yes. It reads your live Safari session, so any page you are already signed into — internal dashboards, paywalled articles you subscribe to — is readable without copy-paste.

Is the page content sent to a server?

No. The text is read locally on your Mac and handed directly to your AI client. LMCP keeps no cloud copy and stores nothing on a server.

Read a Safari Tab With AI on Mac

safari_read_tab pulls the rendered text content out of an open Safari tab and hands it to your AI assistant. You point at the tab by its URL (a partial url_match works) or by its index, and the AI gets the actual readable text of the page — headlines, body copy, lists — not raw HTML. A natural prompt looks like: "Read the GitHub release notes tab open in Safari and summarize what changed in three bullets."

Because it reads the rendered content, it captures what you actually see on screen, including text loaded after the page rendered. That makes it ideal for articles, documentation, dashboards, release notes, and long threads you already have open. It is a read-only tool, so it never clicks, types, or changes anything on the page. Want to try it now? Download LMCP and connect your AI in a couple of minutes.

Which AI agents work?

LMCP exposes safari_read_tab to every major AI client. Desktop apps auto-configure over a local stdio connection, and web AIs connect through the LMCP Cloud Relay connector.

Desktop (auto-configure via local stdio): Claude Desktop, Cursor, VS Code (GitHub Copilot), Windsurf, and Zed.
Web (via LMCP Cloud Relay connector): ChatGPT, Claude.ai, Grok, and Perplexity.

The same tool, and the rest of the 150+ tools, become available regardless of which assistant you prefer.

Automation

safari_read_tab is most powerful when chained with other LMCP tools. Read a tab, then have the AI act on it:

Read a research article in Safari, then create_note in Notes with a tidy summary.
Read a product page, then compose_email to send a colleague the highlights.
Read a meeting agenda tab, then create_reminder for each action item.
List open tabs, pick the right one by url_match, read it, then create_event from the dates inside.

The AI orchestrates these steps in one request, so "read this tab and turn it into a task list" becomes a single instruction.

Context

This tool works against your real Mac and your real Safari session. That means the AI sees pages behind your logins, your internal dashboards, and documents only your account can reach — without you copying and pasting anything. The AI is grounded in what is actually open in front of you, not a generic web fetch that might hit a paywall or a different version of the page.

Productivity

Instead of selecting all the text on a page, copying it, switching apps, and pasting it into a chat, you just say "read the tab." For a long article or a dense doc that is 30 seconds to a minute saved every time, and it scales: ask the AI to read five open tabs and compare them, and you skip a tedious round of manual copy-paste entirely. Reviewers, researchers, and anyone juggling a dozen tabs get the biggest lift.

Privacy & GDPR

Everything runs locally on your Mac. The page text is read on-device and passed straight to your AI client — LMCP keeps no server-side copy, requires no API keys, and stores nothing in the cloud. There is no LMCP database recording which tabs you read. This local-first design makes LMCP GDPR-compliant by architecture: the data never leaves the device you control.