This guide changed in July 2026
The original version of this page connected Google Ads to Claude through Zapier, because in April 2025 there was no official option.
There is now. Google publishes its own open-source Google Ads MCP server, and it has been maintained since October 2025. It is free, and it is the route this guide now covers.
The one thing to know before you start: it is strictly read-only. More on what that means below, because it is the most important fact on this page and almost nobody states it.
In this article, we’re going to show you to connect your Google ads account directly to Claude using MCP, so that you can chat with the data without having to even sign in to Google ads!
Usually, AI tools like Claude and ChatGPT can only work with what you paste into them. MCP lets an assistant reach your actual account data, so you stop copying reports into a chat window.
MCP (Model Context Protocol) is the standard that connects AI assistants like Claude or ChatGPT to outside services. It solves a real limitation: these tools are good at reasoning over information, but on their own they cannot see your account.
With Google’s server, “reach your data” means read it and nothing more. That is a deliberate design choice and it is covered properly below.
P.S. We wrote another guide on Facebook Ads MCP to do the same thing.
Imagine asking Claude questions like…
- “Which campaigns had the highest ROI last month?”
- “Show me ads with declining CTR”
- “Analyze my search term performance across all campaigns”
This is what we’re going to do!
The most important thing about Google’s server
Google’s own documentation is blunt about this:
This implementation is strictly read-only. It cannot modify bids, pause campaigns, or create new assets.
It exposes three tools: list_accessible_accounts, search to run a GAQL query, and get_resource_metadata to describe what fields exist. Alongside them sit four read-only resources, including the API discovery document and the release notes. Every one of them answers a question. None of them changes anything.
The project has been open source since 3 October 2025 under Apache-2.0 and is still moving: the most recent change at the time of writing was a July 2026 security fix stopping OAuth credentials being written to logs, contributed from outside Google.
So an assistant connected to this can tell you your Brand campaign’s CPA tripled last Tuesday. It cannot then pause it.
This is a feature, not a shortcoming. The worst a read-only server can do is tell you something wrong, and you can check that. A write-enabled server that gets something wrong spends your money and you find out from the invoice. If you want the write-enabled comparison, Meta’s server is the opposite design and we compared the two directly .
What you’ll need
- A Google Ads account with administrative access
- A Google Ads developer token. This is the part every “5 minute setup” guide leaves out.
- Python, and a terminal you are willing to use
- Claude Code, or another MCP client that can run a local server
The developer token is the real barrier
A developer token means an application through the API Centre against a manager account, and approval is not instant. Google publishes the timelines: Basic Access is reviewed within five business days, Standard Access within ten .
So if you already have a token you are about ten minutes away. If you do not, the honest answer is up to two working weeks, not an afternoon. Start the application before you read the rest of this page.
One more thing the setup guides skip: the token has to come from a real manager account, not a test one. Google’s documentation is explicit that it “cannot be a Google Ads test manager account”, which is the exact wrong turn that costs people a second application.
Setting it up
The server runs on your own machine, which matters for a reason that trips people up: Claude Desktop’s connector feature reaches out from Anthropic’s cloud, so it cannot see a server running on your laptop. Claude Code is the natural home for this one.
Run Google's official MCP server
pipx run --spec git+https://github.com/googleads/google-ads-mcp.git google-ads-mcpYou will need GOOGLE_PROJECT_ID and GOOGLE_ADS_DEVELOPER_TOKEN set in the environment, plus an OAuth client. Google’s docs cover the Docker and Cloud Run alternatives if you would rather not run it locally.
Then register it with claude mcp add and point it at the command above.
When it has worked, the connector shows as Connected and your assistant can list the accounts it can reach. If it lists nothing, the token is the first thing to check, not the config.
What to ask it once it works
Start with questions that need real data and would take you twenty minutes in the interface. Anything you could answer from memory is a waste of the connection.
That is the shape of a useful answer: named keywords, real spend, zero conversions, and a second list of what is working so you do not cut the wrong thing.
We also made our own internal list of PPC prompts 100% free and public .
Also, since Claude is elite at coding, one of our favorite exercises is to firstly have Claude audit our Google ads account, and then ask Claude what specific Google ad scripts it suggests for our account.
Not only that, but once you have a list of potential scripts to run on your account, you can even have Claude write the script code out for you and later test and debug.
Should you use this at all?
Be honest about the job you are doing.
Use it if you want an analyst: something that answers real questions from real account data instead of guessing. That is what it is good at, and the read-only design means it cannot hurt you.
Do not use it if what you actually want is “tell me what is wrong with my account”. You would be installing a data pipe and then still needing to know which questions to ask. Our audit agent reads the account and hands back findings , with no MCP wiring at all.
If you run Meta as well, its server works in a completely different way: hosted, no developer token, and it can write. One reads, one writes covers which one to trust with an assistant, and the three routes for Meta covers the easier options there.
- Is the Google Ads MCP server free?
- Yes. It is open source under Apache-2.0 and Google publishes it at github.com/googleads/google-ads-mcp. You run it yourself, so the only cost is whatever your AI assistant subscription costs and whatever the machine or container running it costs. There is no per-query charge and no beta pricing to worry about.
- Do I need to know Python to use it?
- You need to be able to run one command in a terminal and set two environment variables. You do not need to read or write Python. If a terminal is genuinely a blocker, Google's docs cover running it in Docker or on Cloud Run instead, which moves the same problem somewhere with a UI.
- Why can't I add it to Claude Desktop as a connector?
- Because connectors reach out from Anthropic's cloud to a public URL, and this server runs on your own machine where that cloud cannot see it. That is a consequence of the local-first design, not a bug. Claude Code runs the server locally and talks to it directly, which is why it is the natural client for this one.
- How long does the developer token really take?
- Google publishes five business days for Basic Access and ten for Standard Access. Basic Access is enough for the read-only queries this server makes. The most common cause of a rejection or a delay is applying from a test manager account, which Google's documentation explicitly rules out.
- Can I use it across a whole MCC?
- Yes. The list_accessible_accounts tool returns every account your credentials can reach, and you set the login customer ID to the manager account while querying a child account for metrics. Asking a manager account directly for performance metrics returns an error rather than data, which catches most people out once.
- What can it not tell me?
- Anything that is not in Google Ads. It cannot see your CRM, your actual revenue, your landing pages or your Meta spend, so it cannot work out true ROAS or tell you whether a lead was any good. It also cannot act: no pausing, no bid changes, no new assets.