Skip to content

Google Ads MCP: what it is and how to connect it safely

AI and toolsUpdated 14 Nov 20264 min read

Cover: an AI app connected to Google Ads through an approval gate in the middle
In this article
  1. What MCP is
  2. Before you connect: what to decide
  3. Three addresses, three levels
  4. How a change reaches Google Ads
  5. There's no token to paste
  6. Connect it to Claude
  7. Checklist before you let an AI touch ad accounts
  8. Try it
  9. Sources

If you already ask Claude or ChatGPT about your campaigns by pasting screenshots and exports, an MCP connection lets the AI read your Google Ads accounts directly. This article explains what MCP is, how to connect AdPilot for Google Ads to your AI app, and how to decide how far the AI is allowed to go.

What MCP is

MCP (Model Context Protocol) "is an open-source standard for connecting AI applications to external systems" (source). Through it, AI applications like Claude or ChatGPT can connect to data sources, tools and workflows (source).

A "Google Ads MCP" is an MCP server whose tools work on Google Ads accounts: reading campaigns, keywords and spend, and — depending on the server — preparing changes.

Before you connect: what to decide

Connecting an AI to an ad account means deciding two things up front:

  • How far the AI may go. Only read? Read and prepare changes? Or also apply them?
  • Who approves a change before it reaches Google Ads.

AdPilot answers both with addresses and an approval gate.

Three addresses, three levels

AdPilot for Google Ads has three addresses on one server; they differ in how far the AI app is allowed to go (source):

AddressWhat the AI can do
FullRead, create proposals, and apply. High-risk changes still need a person to approve them on the web.
View + proposeRead and create proposals; the AI can't apply anything on its own.
View onlyThe AI can see your accounts, but can't propose or change anything.

Still uneasy? Paste the View only address — the AI can look, but can’t propose or change anything (source).

The real level for each call is the lowest of four things: the permission you gave the AI app, the address you're using, your organisation's maximum level in Guardrails, and the plan you're on (source). So if your organisation sets its Guardrails maximum below the address someone pasted, the Guardrails level wins.

How a change reaches Google Ads

AdPilot doesn't write straight to Google Ads. A change becomes a proposal, with its reason and risk level, checked against safety rules and the limits you set. High-risk changes require a person to open the approval page and click; the AI can't approve on its own. Each apply is logged — who approved, when, and the result — and changes can be undone (source).

There's no token to paste

Everyone pastes the same address; it carries nothing that's yours. After you paste it, the AI app opens ALFA's sign-in screen, you sign in and pick your organisation. Access is tied to your account, not to a string in a link, so you can revoke each app separately, and the log knows who called what (source).

Connect it to Claude

  1. Open Customize → Connectors from the left sidebar on Claude.ai (claude.ai/customize/connectors).
  2. Click + Add, then Add custom connector.
  3. Type a name yourself, for example ALFA AdPilot Google — Claude doesn't take the name from the server.
  4. Paste one of the three addresses.
  5. Connect, then sign in to ALFA and pick your organisation.

(source) For ChatGPT, the steps differ (server URL, OAuth, and a "high risk" confirmation box); follow the ChatGPT steps on the connection page.

ALFA has tested the connection on Claude.ai, Claude for desktop and ChatGPT (26 Sep 2026) (source).

Two practical notes (source):

  • The MCP route runs inside your own AI app, so it follows that app's terms — some apps only allow custom connectors on a paid plan. No paid AI plan? Use Studio on the ALFA web app; ALFA pays for the tokens.
  • If AdPilot adds new tools, they only show up after you reconnect.

Checklist before you let an AI touch ad accounts

  • Picked the address for how far the AI may go
  • Set your organisation's maximum level in Guardrails
  • Know who approves high-risk changes on the web
  • Know where the log is, and how to undo a change
  • Know how to revoke the AI app's access

Try it

See how to connect AdPilot to your AI · AdPilot for Google Ads

Running Meta ads too? See AdPilot for Meta Ads.

💬 Need help? Join the ALFA Software support group on Telegram (ask in your own language) — or message our support bot privately about your account or invoices.

Checked on 7 Nov 2026 against modelcontextprotocol.io and ALFA's connection page.

Sources

  1. Model Context Protocol — What is the Model Context Protocol (MCP)?: https://modelcontextprotocol.io/docs/getting-started/intro
  2. ALFA Software — Connect AdPilot for Google & Meta Ads to AI: https://alfasoftwares.com/en/noi-ai
  3. ALFA Software — AdPilot for Google Ads: https://alfasoftwares.com/en/adpilot-google