Google Ads MCP server: connect your AI to real ad accounts

What the MCP server does

MCP defines how an AI application discovers and calls tools. Adport connects those tools to advertising providers. Your assistant supplies the analysis and conversation; Adport supplies account access, normalized reports, and the policy checks around changes.

Connecting Adport does not move your campaigns to a new advertising platform. Google remains the source of the ads, billing, and platform permissions. The connection makes supported operations available through a common interface. Your AI subscription, advertising spend, and Adport plan are separate.

Connect Google and choose the right project

1. Create or select the Adport project for the client or brand you want to work on.
2. Open Accounts, then Platforms, and connect Google through its authorization flow.
3. Save the specific ad accounts you want available in this project.
4. Enable agent access for the accounts you intend to use.
5. Open Agent access and copy the hosted MCP address.

For agencies, use a separate project for each client. Do not authorize every account just because your Google identity can see it. An account that is discoverable through a manager account still needs the right underlying Google access. Google documents the difference between manager access and administrative ownership.

Add the server to your AI assistant

The hosted endpoint is:

https://app.adport.dev/mcp

Use the client-specific instructions for Claude, ChatGPT, or Codex. Choose OAuth where the client offers an authentication choice. Sign in to Adport and select the project the connection should access.

An existing agent authorization stays bound to its project. Changing the project in your dashboard is not a way to redirect an already-authorized agent to another client. Authorize the appropriate project in the agent when you need a different scope. Keep separate connections clearly named when your client supports multiple server entries.

Run a small, read-only check first

Start with an account inventory, then request a short campaign report. Confirm one known account ID and one campaign before trusting a portfolio-wide answer. Ask the assistant to preserve each account's currency and to label partial results.

A useful first answer includes its period, the included accounts, and any provider warnings. A list of plausible campaign names is not enough. If the account is missing, return to account selection rather than asking the model to guess an ID.

Move from reporting to a reviewed change

After the read check, ask for a narrowly scoped proposal. Give the exact account and campaign, the intended outcome, and any budget ceiling. Adport's first write call returns a preview and a pending operation. Applying that operation requires the matching token and identical arguments before expiry.

Review the actual change, not just the assistant's explanation. Check account identity, the old and new values, currency, any policy adjustments, and whether provider-side validation was available. A changed argument or an expired preview needs a fresh preview. New campaigns are subject to the project's paused-creation policy.

A Free cloud plan is read-only. Paid write access still depends on the member's role, enabled accounts, and the provider's permissions. Upgrading the plan does not override those boundaries.

If the connection is empty or fails

• No tools: check that the client enabled the MCP server and completed authentication.
• No accounts: verify the selected Adport project, saved Google account selection, and agent-access switches.
• Reads work, writes fail: check role, plan, protected accounts, and budget limits.
• Expired authorization: reconnect through the client instead of pasting a platform token into chat.
• Partial report: resolve the named provider error before treating the answer as a complete portfolio view.

Once the connection is verified, use the Google Ads audit checklist to turn access into a repeatable review.