What AI Agents Can Do

Once connected over MCP, an AI assistant gets a focused set of capabilities over your cloud workspace. This page explains them by use case — every capability comes with an ask you can copy and send. The MCP tool reference lists every tool by name, and Prompts to try is the full cookbook.

Understand your workspace

The agent can read your plan, credit balance, and its own access level — so it can answer "how many credits do I have left?" and warn you before starting anything expensive.

Work with your data

• Browse tables — list all result tables, grouped by the automation that produced them.
• Inspect — read a table's schema, row count, and a small sample.
• Query without limits — filter, aggregate, group, and sort tables server-side. The full table is never loaded into the AI's context, so a 100,000-row table is as easy to analyze as a 100-row one.
• Export — generate CSV, JSON, or Excel downloads, or send a table straight to Google Sheets (Sheets needs a full-access connection).

Clean your data

With a full-access connection, agents can tidy tables the way you would by hand:

• Rename columns or change their type
• Delete irrelevant columns
• Merge 2–5 columns into one (street + city + zip → Address)
• Delete rows that match a filter — "drop every row with no price or image"
• Remove duplicate rows, keeping the first occurrence

Row deletion is never silent: the agent always gets the match count first and deletes only after you confirm.

Operate your automations

• List automations and their run history — including "what ran this week?" (available to every connection, including read-only)
• Inspect any automation in depth: its full configuration, schedule state, recent runs, and latest results table
• Check a run's live progress — and on a failed run, get the reason, how many rows were kept, and what to try next
• Start or stop a run (full access) — starting shows a cost estimate from the automation's own run history first

This is what makes "re-run my competitor price scrape and put results in Sheets" a one-sentence job — including for point-and-click scrapers you built in the extension and run in the cloud.

Manage your automations too, with full access: put one on a schedule, change or clear that schedule, pause and resume it, rename it, or delete it. Pausing keeps the schedule in place for when you resume; deleting — always confirmed with you first — removes the automation and its schedule, but the data tables it produced stay.

Start new extractions

With a full-access connection, the agent can also go out and get new data — from pasted URLs, from a URL column of an existing table, or from URLs it discovers itself:

• Scrape similar pages of one site — products, listings, articles, job posts, real-estate pages — one row per URL, structured data captured automatically, no selectors. Page metadata and phone numbers can ride along as extra columns.
• Build lead lists across many sites — emails and social profiles, one row per site, with an optional deep scan of each site's contact and team pages.
• Extract map place pages — name, rating, review count, address, phone, website, and opening hours, one row per place.
• URL discovery — mapping a whole site from its sitemaps, so "just the winter collection" becomes a URL table with no pasting.
• Whole Shopify catalogues — extracted as clean per-variant product tables.

This is powerful enough to get its own page.

Get product help

The agent has this documentation at its fingertips on every connection, including read-only: it can search and read every docs page — the extension tools, the Cloud Platform, credits, guides — and answer questions from what the docs actually say, with a link to the right page.

Access levels

Choose read-only when you want an analyst; choose full access when you want an operator. You pick the level when you connect — on the consent screen for sign-in apps, on the token for everything else.

Related