The official Google Ads MCP server: your account, in an agent's hands

An MCP server is a small bridge that hands an AI agent a fixed set of callable tools instead of leaving it to guess at an API. This one is Google's own bridge into the Google Ads API, and it closes a gap every agent builder on this beat has hit: there was no first-party way to let an agent look inside a real account without hand-rolling the auth and query plumbing yourself.

Why it made the cut

It exposes the tools an agent actually needs to orient itself in an account: search across accounts, list every customer a set of credentials can reach, and fetch resource metadata straight from the Google Ads API. Because Google maintains it, the tool surface tracks the real API rather than a third party's interpretation of it, and it is announced on the Ads Developer Blog rather than surfacing quietly on GitHub, which is as close to an official endorsement as this category gets.

How to use it

Clone the repository, set up Google Ads API credentials as you would for any API integration, and point an MCP client such as Claude or Cursor at the server. It supports an OAuth proxy for dynamic user authentication, so a single deployment can serve multiple users without baking one set of credentials into the server itself, useful if you are building this into a shared internal tool rather than a solo setup. Once connected, the agent can search accounts and pull resource data on request instead of you writing GAQL by hand for every question.

Worth knowing

It is free and Apache-2.0 licensed, and you self-host it yourself: there is no hosted version to sign up for, and normal Google Ads API usage costs still apply behind it. It is built on FastMCP. Pair it with the house Google Ads audit Skill on this shelf: the MCP server is how the agent fetches the data, the Skill is how it reasons over what comes back.