A Google Ads MCP server you don't have to run

What is a Google Ads MCP server?

A Google Ads MCP server is a program that sits between an AI client and the Google Ads API and presents the API as a list of tools the model is allowed to call. Ask your assistant “which search terms spent more than fifty dollars last month without a conversion?” and it calls a search tool, gets structured rows back, and answers from them. Ask it to add those terms as negatives and — if the server offers write tools and you have enabled them — it calls add_negative_keywords.

The important word is tools. The model is not screen-scraping your dashboard or guessing at an export. It is calling the same API the Google Ads interface calls, with the same permissions your Google account already has, and getting the same numbers back.

What is MCP, in one paragraph?

MCP is the Model Context Protocol, an open standard published by Anthropic in late 2024 and since adopted well beyond it. It standardises how an AI client discovers and calls external tools, so one server can serve many clients. Practically, it means a Google Ads connector written once works in Claude, ChatGPT, Copilot, Gemini CLI, Cursor and anything else that speaks the protocol — and it means the connector is a URL you paste in, not a plugin you install.

Google's official server versus a hosted one

Google maintains an open-source Google Ads MCP server at github.com/googleads/google-ads-mcp, Apache 2.0 licensed. It is a good piece of software and it is genuinely free. It is also read-only, and you host it yourself. AdCopilot is a fork of it: the same reading core, extended with write tools, and run for you.

Which tools does the AdCopilot server expose?

Thirty-six tools exist in the underlying server. Thirty-two are exposed to hosted connectors, in seven groups. On the read-only tier only the first group is available.

What it deliberately cannot do

Four tools in the underlying server delete things: remove_ad, remove_keyword, remove_campaign_asset and remove_asset_from_asset_group. None of them is exposed to a hosted connector — not gated behind a setting, not available on request. They are filtered out of the tool list before the model ever sees them.

The reasoning is that deletion is the one class of change that a conversation cannot undo. Pausing an ad is reversible in a click; removing it is not. So the agent can pause, adjust, add and negate, and anything that destroys is left to the Google Ads interface, where there is a confirmation dialog and a human hand on it.

How to connect it

1. Sign in with Google and grant access to the accounts you choose. AdCopilot uses Google's own sign-in, so it runs under your login with the permissions you already have — no password is shared with us.
2. You get a private connector address of the form https://mcp.adcopilot.cloud/c/<your key>/mcp. Treat it like a password: anyone holding it holds your access.
3. Paste it into your AI client as a custom MCP connector. Step-by-step for Claude and ChatGPT.
4. Ask it something. A good first question is one you already know the answer to.

{"mcpServers": {"adcopilot": {"httpUrl": "https://mcp.adcopilot.cloud/c/<your key>/mcp"}}}

What you need

• A Google account with access to the ads account. The connector can reach exactly what Google already lets you reach, and nothing more.
• An AI client you already pay for. The reasoning happens in your assistant, on your subscription. We add no model of our own. (Included AI credits, for people who would rather not bring one, are on the roadmap.)
• No developer token, Cloud project or OAuth client of your own — those are the parts a hosted server exists to absorb.

Frequently asked questions