Connect Google Ads to ChatGPT and Claude

Connecting Google Ads to ChatGPT and Claude takes five ordered steps and about ten minutes: start a pilot, copy your private connector address, add it in the client, sign in with your own Google account, and run a first command. The same address works in both apps, and it runs the full Google Ads toolset from plain-language prompts. One thing to set expectations up front — this executes on your command with an approval on every write, not on a schedule while you sleep.

Step 1 — Start a free pilot and get your connector address

Go to adcopilot.cloud and start the free seven-day pilot. No card, up to five accounts. Signup gives you a private connector address — a URL unique to you that looks like https://mcp.adcopilot.cloud/c/<your connector key>/mcp. Treat this like a credential: it is how the client reaches your connector, so keep it out of shared repos and screenshots. Copy it; every later step pastes this same address.

Step 2 — Add the connector in your client

Both ChatGPT and Claude accept remote MCP connectors in a few clicks. The address from Step 1 is all either needs.

• In ChatGPT: open the connectors settings, choose to add a connector, and paste your address as the remote MCP server URL. The exact menu path moves as the app updates, so the current click-by-click is kept on the ChatGPT setup page.
• In Claude: open Settings, go to Connectors, add a custom connector, and paste the same address. Step-by-step with current screenshots lives on the Claude setup page.

You do not need a different address per app — paste the identical URL in each. If you use other clients too, the same connector works across every MCP client.

A note on Microsoft Copilot: adding an arbitrary remote MCP connector the one-click way that Claude and ChatGPT support is not how consumer Copilot works today. MCP reaches Copilot through Copilot Studio and Microsoft 365 Copilot agents, which is still rolling out — so if Copilot is your goal, plan for that path rather than expecting parity with the two clients above.

Step 3 — Sign in with your Google account and authorize scope

The first time the connector runs, it walks you through Google sign-in. Use your own Google account — the one with access to the ad accounts you manage. Google's consent screen shows the read/write scope the connector requests; approve it there. This is standard Google OAuth: you are granting access under your own login, and you can revoke it from your Google account security settings at any time. Nothing is stored as a shared password.

Once authorized, the connector can see the accounts your Google login can see — the same list you would find in the Google Ads UI.

Step 4 — Run a first read command

Confirm the connection with a read before you change anything. In the chat, ask:

List the Google Ads accounts you can see.

Then a real reporting question against one account:

In account <account>, what did we spend in the last 7 days, by campaign,
with conversions and cost per conversion? Just show me, don't change anything.

These are reads — no approval needed, no risk, immediate answers. Check the numbers against the UI once. That quick calibration is what makes you trust the approvals that come next, because you have seen the connector read your account correctly.

Step 5 — Make one approved change

Now ask for a change. Something small and reversible is the right first move:

In account <account>, add "free" and "cheap" as negative keywords to the
Brand campaign.

The client pauses and shows you the write it is about to make — the tool, the account, the exact values — and waits. Nothing happens until you approve. That per-write confirmation is the core safety model: the connector proposes, you approve, then it executes. The sane policy is to graduate boring reads to always-allow while keeping budgets, statuses, and anything spend-affecting on manual approval every time. There is a fuller approval workflow guide for teams.

What it can and cannot do

Two honest boundaries worth internalizing:

• It runs on your command. It acts when you ask, inside a session, with an approval on every write. It does not run overnight or on a schedule in the shipped product. Scheduled audits and rolling reports are being trialed with pilot accounts, not generally available.
• It can never delete. The four destructive remove tools are never exposed to the AI, and a removal instruction is refused server-side. It can pause, adjust, and create across 36 tools — 32 full, 3 read-only — but there is no delete path for it to take.

You're connected

That is the whole flow: pilot, address, add in ChatGPT or Claude, Google sign-in, first read, first approved change. From here you drive the account in plain language and approve each write as it comes. AdCopilot is built by Atromx Intelligence and is not affiliated with Google, Anthropic, or OpenAI — you sign in under your own Google account and stay in control of every change.