What is an MCP server for Google Ads?
An MCP server for Google Ads lets an AI assistant read reports and change campaigns through the Google Ads API with your authorization. Here is how it works, how it compares with Google Ads Scripts, the Editor and BI dashboards, and what to check before trusting one.
The one-sentence definition
An MCP server for Google Ads is a service that exposes Google Ads operations as tools an AI assistant can call, using the Model Context Protocol, with the advertiser's own Google authorization.
MCP (Model Context Protocol) is an open standard released by Anthropic in late 2024 and since adopted by OpenAI, Google, Microsoft, Cursor and most AI IDEs. It defines how an assistant discovers tools, calls them with structured arguments and receives results. A "server" is anything that implements that contract. PaidHelm is one for Google Ads.
How a call travels
- You type: "Which campaigns are pacing above budget this month?"
- The assistant chooses a tool, here
consultar_relatorio, and writes a GAQL query for cost and budget by campaign. - The MCP server validates the request, checks your plan quota, and calls the Google Ads API with your OAuth token.
- Google returns rows; the server returns them to the assistant; the assistant does the arithmetic and writes the answer.
Writes work the same way, with one difference: a well-designed server makes the assistant obtain IDs from a read first and creates new campaigns paused, so nothing goes live by accident.
Versus Scripts, Editor and dashboards
| Google Ads Scripts | Google Ads Editor | BI dashboard | MCP server | |
|---|---|---|---|---|
| Who operates it | Someone who writes JavaScript | Anyone, manually | Analyst builds, everyone reads | Anyone, in plain language |
| Reads | Yes, scheduled | Yes, bulk | Yes, visual | Yes, any question, on demand |
| Writes | Yes, pre-programmed | Yes, bulk, manual | No | Yes, with approval |
| Handles a new question | Write a new script | Click through | Build a new view | Just ask |
| Reasoning over results | None | None | None | The assistant explains and recommends |
Scripts are still the best choice for a fixed rule that must run every hour unattended. Editor remains excellent for a thousand-row bulk edit. Dashboards are for stakeholders who only read. The MCP server covers the daily reality in between: a human with a question, a decision to make and a change to apply.
What to check before trusting a server
- Authorization model. It must use Google OAuth with the Google Ads scope only. Never type a Google password or share a developer token with a service you do not control.
- Who holds the developer token. Hosted servers use their own approved token; open-source servers need yours, which means applying for API access and keeping a server running.
- Default state of new objects. Paused by default or live by default? Ask.
- Data retention. Does the server store your reports and campaigns, or only relay them?
- Revocation. You should be able to cut access in one click in Google and in the product.
- Quotas and pricing. Understand what a billable unit is. PaidHelm counts tool calls as tasks; the counter is visible in the chat.
Getting started
With a hosted server the whole setup is: pick an assistant, paste the server URL, sign in with Google, ask a question. The PaidHelm quickstart walks through it in five minutes on the free plan.
Frequently asked questions
Is an MCP server the same as a Google Ads plugin?
Do I need to code to use one?
Put PaidHelm on your own account.
Free plan, no credit card. Connect in five minutes and ask for your first report.