The full setup.

This is the part that does the heavy lifting, and you only do it once. You link Claude to Apify through a connector. After that, Claude can run any Apify scraper for you just by naming it. No dashboards, no code.

In Claude, open Settings → Connectors → Add custom connector and enter these two values:

Click Add. Apify will ask you to verify and sign in, which links your Apify account to Claude. Approve it. (No API token to copy, the sign in handles it.)

That’s the whole setup. From here, it’s just conversation.

Before the real run, make sure Claude can actually reach Apify. Paste this. It only checks the connection, so it doesn’t run anything or cost a cent.

Quick check: are you connected to Apify? Confirm by listing the
Apify tools you can access right now. Do not run anything yet.

If Claude lists its Apify tools, the connection is live and you’re ready. If not, reopen the connector in Settings and reconnect.

Now point Claude at the exact scraper. Open the Google Maps Scraper on Apify (or search the Apify Store for Google Maps Scraper by Compass). Apify calls each scraper an actor, so that’s the word you’ll see on its page.

On the actor page, click the API button (top right), then choose MCP configurator.

That opens the scraper’s MCP server URL. Copy it.

Then paste that URL into Claude together with this prompt. It checks that Claude can reach and query the scraper before you run anything. (The URL is already included below.)

Here is the scraper I want to use:
https://mcp.apify.com/?tools=actors,docs,compass/crawler-google-places

Can you access this actor and confirm you can query it? Do not run a full job yet,
just check that you can reach it. If you cannot, tell me the exact error.

Now run it. Paste this. Claude asks what to search, then pulls a real sample so you see exactly what comes back.

Let's run a small live test. First, ask me what to search: the type of
business and the location. After I answer, run the Google Maps scraper for 5
businesses, and for each one pull up to 10 reviews sorted lowest rating first so
any complaints show up. Then show me the complete raw data for one business, every
field, without skipping or summarizing, and confirm the review text is actually there.

This is the real run, not a throwaway demo. The reviews are live, and the next step turns them straight into your brief. To audit specific competitors, paste their Google Maps links instead of a business type.

Once Claude has the reviews, give it the analysis prompt. It reads all of them and writes the one page brief, ranked by how often each theme shows up.

You now have the reviews. Produce a one page brief from only what's in the
data. Do not invent counts or complaints that aren't there.

1. The things customers love most, each with how many reviews mention it.
2. The biggest complaints, each with how many reviews mention it. If there
   are none, say "no complaints found" instead of forcing the list.
3. Every recurring theme ranked by frequency, each with a verdict
   (loved / mixed / complaint).
4. Up to three gaps a competitor could win on, only if the reviews support
   them. If they don't, say so.
5. Three specific actions to take next.

Quote one short real review per theme. Flag any reviews that look seeded or
fake.

Then create two actual downloadable files, not text in the chat:
1. The full brief as a downloadable document.
2. The table (columns: Business, Rating, Top complaint) as a downloadable
   CSV I can open in Google Sheets.

Give me a download link for each file.

Claude hands you two downloads: the brief as a document and the table as a CSV. Open the CSV in Google Sheets to sort and track it, and keep the brief to share. Click Download on each, or grab both with Download all.