Google Ads MCP: Connect Claude to Google Ads, Step by Step

The Google Ads MCP server is Google's official way to let an AI assistant, like Claude, read your Google Ads data. You run it on your computer or in Google Cloud, connect it to your AI tool, and then ask questions like "how did my campaigns do this week" in plain English. Right now it is read-only: it can look, it cannot change anything.

It is free and it works. It is also a developer tool. You need a Google Cloud project, API access and credentials before the first question. This guide covers what it does, how to connect it to Claude, and when a no-setup option makes more sense.

What MCP is, in plain words

MCP stands for Model Context Protocol. The project's own site calls it "an open-source standard for connecting AI applications to external systems", and compares it to "a USB-C port for AI applications."

Without MCP, an AI chat only knows what you paste into it. With an MCP server, the AI can call tools to fetch live data itself. For Google Ads, that means it can pull your real campaign numbers instead of guessing from a screenshot.

What the official Google Ads MCP server does

Google's developer guide describes it as "a standardized bridge to the Google Ads API, which lets AI agents analyze and retrieve campaign data using natural language."

The key facts, from Google's guide:

It gives the AI three tools:

• list_accessible_customers - lists the Google Ads accounts your login can reach.
• search - runs Google Ads Query Language (GAQL) requests to fetch metrics, budgets and status. GAQL is a short query language, a bit like asking a spreadsheet for exact rows and columns.
• get_resource_metadata - explains what fields a resource like "campaign" has, so the AI can write correct queries.

The GitHub instructions add resources the AI can read: the API's discovery document, the lists of metrics and segments, and the latest release notes.

What it cannot do

Google is direct about it: "This implementation is strictly read-only. It cannot modify bids, pause campaigns, or create new assets."

So you can ask "which campaigns spent the most last week, split by device", but not "pause the worst one". For many people that is a feature. An AI that cannot spend your money cannot make an expensive mistake with it.

What you need before you start

From Google's guide and its GitHub instructions:

1. A Google Cloud project with the Google Ads API enabled.
2. API access for that project at Explorer, Basic or Standard level. Apply if you do not have it.
3. Credentials: an OAuth 2.0 client ID and secret, or Application Default Credentials.
4. Python tooling: the GitHub instructions use pipx to run the server.
5. An MCP client, like Claude Desktop, Claude Code, Gemini CLI or another MCP-compatible tool.

A note on developer tokens, because older guides all mention one. Google's developer guide, checked on 2026-10-08, says developer tokens "were sunset on September 9, 2026, and are no longer required in the latest version of the Google Ads MCP server because API access levels are associated with the Google Cloud project." The GitHub instructions now mark the developer token step as optional.

If you reach accounts through a manager account (MCC), set GOOGLE_ADS_LOGIN_CUSTOMER_ID to the manager's customer ID.

How to connect the Google Ads MCP server to Claude

Option 1: Claude Desktop

The MCP project's guide for Claude Desktop:

1. Open Claude Desktop Settings from the Claude menu in your system menu bar (not the settings inside the chat window).
2. Go to the Developer tab and select Edit Config.
3. This opens claude_desktop_config.json. On macOS it lives in ~/Library/Application Support/Claude/, on Windows in %APPDATA%\Claude\.
4. Add the Google Ads server. Google's example from GitHub looks like this:

{
  "mcpServers": {
    "google-ads-mcp": {
      "command": "pipx",
      "args": [
        "run",
        "--spec",
        "git+https://github.com/googleads/google-ads-mcp.git",
        "google-ads-mcp"
      ],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "PATH_TO_CREDENTIALS_JSON",
        "GOOGLE_PROJECT_ID": "YOUR_PROJECT_ID"
      }
    }
  }
}

5. Save the file and restart Claude Desktop.

The MCP guide's troubleshooting tips: check the JSON syntax, and use absolute file paths, not relative ones.

Option 2: Claude Code

Claude Code adds local servers from the terminal. Its docs show the pattern for a stdio server: claude mcp add, your environment variables with --env, --transport stdio, a name, then -- followed by the command that starts the server. Everything after -- is passed to the server untouched. Use the same pipx command and environment variables as the JSON above.

Option 3: a shared server on Google Cloud Run

If a team wants one server for several people or agents, Google's guide shows how to build the image with Cloud Build and deploy it to Cloud Run. Clients then connect to the Cloud Run URL instead of starting the server themselves. This is the most work, and the most control.

First questions to ask

Google's guide suggests starting with:

• "What can the google-ads-mcp server do?"
• "What customers do I have access to?"
• "How many active campaigns do I have?"
• "How is my campaign performance this week?"

Then get specific. Questions that work well on a read-only connection:

1. Which search terms spent money with no conversions in the last 30 days?
2. Which products in Shopping or Performance Max spent the most with the lowest ROAS?
3. How did cost per click change month over month, by campaign?
4. Which campaigns are limited by budget?

Google Ads MCP vs other ways to put AI on your account

On third-party servers: read what each one can do before you connect it. Some list write actions. One hosted Google Ads MCP page we checked lists actions like setting a campaign's status. A write-capable connection can pause or change a live campaign from a chat message.

For more on the safety side, read whether it is safe to give AI access to Google Ads and whether AI can read your account without changing it.

The limits worth knowing

• It sends your data to the AI you connect. Google's GitHub instructions say so plainly: "The MCP Server will expose your data to the Agent or LLM that you connect to it."
• The AI writes the queries. GAQL is precise. A wrong field or date range gives a confident wrong answer. Check numbers that matter against the Google Ads interface.
• It knows data, not judgment. It can tell you a campaign's ROAS. It does not know your margin, so it cannot tell you whether that ROAS makes money unless you tell it.
• Read-only for now. Google's guide says "current release". That may change, so recheck before you rely on it.

The no-setup option

We build one of the alternatives, so weigh this accordingly. Disclosure: Christopher Krassnig, who founded Scaley, also runs ZenoX Media, a Google Ads agency.

Scaley Media Buyer is the same idea as a read-only Google Ads connection, without the Cloud project, credentials or config files. You connect Google Ads read-only from inside the app and ask questions about your live account. It cannot change a setting. It is also built to answer like a senior media buyer, so you get a read on the numbers, not only the numbers.

If you want changes made too, Scaley Suite suggests them and makes each one only after you approve that change. See how AI fits into Google Ads, compare with the ChatGPT routes, or check whether ChatGPT can manage Google Ads. Prices for all three tiers are on the pricing page, and you can start the free trial.

Where these facts come from

Checked on 2026-10-08:

• Google Ads MCP server: Developer integration guide, Google for Developers - read-only mode, transport, sign-in options, the three tools, prerequisites, the developer token sunset, Cloud Run deployment and sample prompts.
• googleads/google-ads-mcp on GitHub - the setup instructions: tools and resources, the optional developer token step, client configuration examples, the login customer ID, and the data exposure note.
• What is the Model Context Protocol, modelcontextprotocol.io - the definition of MCP.
• Connect to local MCP servers, modelcontextprotocol.io - the Claude Desktop settings path, config file locations and troubleshooting tips.
• Connect Claude Code to tools via MCP, Claude Code Docs - adding a local stdio server with claude mcp add.