Who it’s for

Anyone who checks the same web page on a rhythm — a listings board, a supplier’s prices, a client portal’s table, an internal dashboard — and re-copies the same fields every time because there’s no export and no API.

The pain

The data is right there on the page, but getting it out is manual: open the site, log in, scroll, copy, paste, compare against yesterday. Do that a few times a week and it’s an hour gone — and you still miss the one row that changed.

The recipe

In your AI (Claude, ChatGPT, or Cursor) with LMCP connected, paste:

The prompt — paste into your AI

Open this page in your browser and 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 and you can tell me what's new. Here's the page: <paste the URL>

What the agent does

Opens the page in LMCP’s built-in browser — web_navigate (and web_login first if it’s behind a sign-in; you log in, never the AI).
Pulls the rows into clean fields — web_extract with the columns you asked for, null where a field is missing.
Shows you the table so you confirm it’s the right data before anything is saved.
Saves it as a recipe — recipe_save — so the whole flow becomes one line: “run daily listings.”
Tells you what changed on the next run — recipe_runs keeps the history, so the AI diffs today against yesterday and surfaces only the new, changed, or gone rows.

What you see

A clean table the first time, exactly the columns you named.
Every rerun: a short “here’s what’s new since last time” instead of the whole page again.

Why this is impossible without LMCP

Behind-login pages: a plain scraper can’t sign in as you. LMCP’s browser reuses the session you logged into, on your Mac — so it reads the pages that never had an API.
Repeatable by a sentence: the working pull is saved as a recipe, so “rerun it” is one line — no re-explaining the columns or the URL each time.
Local and yours: the browser, the saved login, and every run’s data stay on your Mac. No scraping service, no API keys, nothing uploaded.

Variations

“…only tell me when a price drops below X” → the agent filters the diff to the rows that crossed your threshold.
“…and drop the table into a Google Sheet each time” → add a write step to the recipe so every run appends the fresh rows.
“…across three sites at once” → save one recipe per page and run them together for a combined “what changed” brief.

Read-only by default. If a recipe ever submits something, it previews first and waits for your confirm. LMCP never solves CAPTCHAs or bypasses bot-detection — if a site hard-blocks the browser, your AI tells you rather than looping.