# What is a Google Ads MCP server?

> How a Google Ads MCP server connects an AI assistant to your ad account, what the options are, and how to use one safely.

A Google Ads MCP server lets an AI assistant such as Claude or ChatGPT read your Google Ads account, and sometimes change it. This guide explains how that works, what your options are, and what to check before you connect a real account.

## What MCP is

MCP stands for [Model Context Protocol](https://modelcontextprotocol.io/docs/getting-started/intro), an open standard for connecting AI applications to outside systems. Its own docs compare it to a USB-C port: one standard way to plug things in.

Without it, an assistant only knows what you paste into the chat. With it, the assistant can call tools that fetch real data or take actions.

## What a Google Ads MCP server does

The server sits between your assistant and the Google Ads API.

1. You ask: "Which campaigns spent the most last week?"
2. The assistant picks a tool from the server, such as one that reads campaign performance.
3. The server fetches the data from your account.
4. The assistant answers using your real numbers.

So you don't need to know the Google Ads interface, or write reports by hand.

## Your options

**Google's official server.** [Google's MCP server](https://developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server) is open source and built for reporting on live account data. Google's docs describe it as read-only: it can't change bids or pause campaigns. Its [repository](https://github.com/googleads/google-ads-mcp) also lists optional tools for assets and labels, so check the version you install. You run it yourself and set up a Google Cloud project and credentials. It suits developers who mainly want reporting, and [Official server or hosted?](/guides/google-ads-mcp-official-vs-hosted) compares the two routes.

**Community servers.** Many open-source servers exist, and some add write tools. Quality and safety vary, you run and update them yourself, and you are trusting that code with your account. Read the code before you connect a real account.

**Hosted servers.** A company runs the server and you connect by signing in, usually with nothing to install. This is the simplest route, and [Roasby](/google-ads) is one of these. The trade-off is that you are trusting that company, so check how it stores your access and what it can do.

## What can go wrong

A connected assistant is useful because it can act, and that is also the risk. Assistants make mistakes, and a misunderstood request can cost real money.

- **Wrong change.** An assistant pauses your best campaign, or changes the wrong budget.
- **Oversized change.** A typo or a misread number raises a budget far more than you meant.
- **Deleted work.** A removed campaign often can't be brought back.
- **Instructions hidden in data.** Text from outside, such as a search term, could try to steer the assistant. This is known as prompt injection, and it is why a person should approve changes.
- **Leaked access.** Keys pasted into a config file or a chat can end up somewhere you didn't intend.

## What to check before you connect

Read access carries little risk to your campaigns, and [Is it safe to give AI access to Google Ads?](/guides/is-it-safe-to-give-ai-access-to-google-ads) covers the wider question. Before you give any server write access, look for these:

| Check | Why it matters |
| --- | --- |
| Changes are previewed | You see what will happen before it does |
| A preview and a confirmation step before each change | The assistant is told to wait for you |
| New campaigns start paused | Nothing spends until you choose |
| No delete tools | A mistake can't remove your campaigns |
| Budget limits sit outside the assistant | The assistant can't talk its way past them |
| Every change is logged, and most are revertible | You can see what happened and usually reverse it |
| Sign-in uses OAuth | No keys to copy, and you can revoke access |
| Access is stored encrypted | A leak is less likely to expose your account |

If a server can't answer these, start with read-only access.

## A safe way to start

1. **Start by reading.** Ask about performance and wasted spend for a few days, and check the answers against Google Ads.
2. **Try a small change.** Add one negative keyword, and see how the preview and the confirmation work.
3. **Set limits first.** Choose a maximum daily budget and a maximum change before you let an assistant touch budgets.
4. **Review the log.** Look at what changed each week until you trust the setup.

## How Roasby does it

[Roasby](/google-ads) is a hosted MCP server for Google Ads, Microsoft Ads and ChatGPT Ads. You add one address to your assistant, sign in, and approve the connection.

It follows the checklist above: your assistant reads freely, previews each change and is told to wait for your yes, new campaigns start paused, and there are no delete tools. The [safety model](/docs/safety-model) has the details, and [Connect your AI](/docs/connect-your-ai) shows the setup for each assistant.