Skip to content

Scrape a Website With AI on Mac, Then Rerun It in One Line

Have an AI scrape a web page — listings, prices, a portal table, even behind a login — then save it as a recipe you rerun with one line, so it tells you what changed since last time. Runs locally with LMCP's built-in browser. No API keys, no scraping service.

L
LMCP··5 min read

You keep going back to the same web page — a listings board, a supplier's prices, a client portal's table, a dashboard — copying the same numbers by hand. There's no export button and no API. LMCP lets an AI do that pull for you on your Mac, and save it so the next run is one sentence. It works through LMCP, a free, native macOS MCP server with a built-in browser — so it can even read pages that live behind a login.

How it works

The pattern is get it working once, then save it as a recipe. LMCP's web tools are agnostic — the same handful work on any site:

  • web_login — if the page needs a sign-in, a real browser window opens and you log in (password, 2FA — never the AI). The session is saved on your Mac and reused, so you don't log in again.
  • web_navigate / web_read — go to the data page and read its content, or a compact list of its elements so the AI can find the right rows.
  • web_extract — pull the rows into clean, structured fields (title, price, link…), null where a field is missing.
  • recipe_save / recipe_run — save the working sequence as a named recipe and replay it anytime with one line. recipe_runs keeps the history, so the AI can tell you what changed since last time instead of dumping the whole table.

Try a prompt like: "Open this page, pull every listing into a table — title, price and link — show me the table, then save it as a recipe called 'daily listings' so I can rerun it tomorrow."

The real value: what changed

A one-off scrape is easy. The point of a recurring one is the diff. Because every run is stored, on the next run your AI compares the fresh rows against the last ones and surfaces only what's new, changed, or gone — the update you actually wanted, not a wall of rows you have to re-read.

Does it run on a schedule by itself?

LMCP doesn't run recipes on an unattended timer — you trigger a run whenever you want it ("run my listings recipe"), or wire it into your own scheduler or an AI routine. That keeps you in control: a scrape happens because you asked, and any recipe that would ever submit something previews first and waits for your confirm.

Is this safe and private?

The browsing happens in LMCP's own browser on your Mac, and when a page needs a login you type the password yourself — the AI never sees it. The saved session, your recipes and the history of every run stay on your Mac. What you ask the AI to read doesn't: the rows it pulls become part of your conversation with your AI provider, under that provider's terms. LMCP executes local tools on your device. Cloud AI providers and connected services can receive requested data; their terms apply.

Which AI assistants can do this?

  • Desktop clients — desktop AI apps connect via local stdio, and LMCP sets up the ones it can automatically.

Stop copying the same page by hand. Download LMCP, point your AI at the page, and save the pull as a recipe you rerun in one line.

Ready to try it?

Works with Claude, Cursor, VS Code, ChatGPT and any MCP client

Download for Mac
FreemacOS 12+ · Apple Silicon & Intel

✦ llms.txt

Comments

Questions and experiences welcome — comments are reviewed before they appear.