Google Ads MCP: Connect an AI Assistant to Your Ads Account
A Google Ads MCP server lets an AI assistant read your campaigns, search terms and spend. What the official server does, how to connect it and which safeguards to set.
02 / In this article
A Google Ads MCP server is a small program that lets an AI assistant query your Google Ads account through the Google Ads API, using the Model Context Protocol. Google publishes an official open-source server, and as of October 2026 its current release is read-only: the assistant can look at campaigns, budgets and search terms, but it cannot change them. This guide explains what the server does, how it connects, which questions are worth asking, and which safeguards to set before you plug an assistant into an account that spends real money.
Key Takeaways
- There is an official server. Google publishes it in the
googleadsGitHub organization and documents it on the Google Ads API site. Its current release is read-only. - It reads, it does not edit. Google's documentation says it cannot modify bids, pause campaigns or create assets. Any server that does write needs much stricter controls.
- Access is the real control. The assistant sees what the signed-in user or service account can see. Give it a read-only user, not an administrator.
- A human approves every change. Budgets, bids, targeting and conversion settings stay with a person, and the assistant only proposes.
- Your data goes to the model you connect. The server's own README warns that it exposes your data to the agent or LLM on the other end.
What is a Google Ads MCP server?
A Google Ads MCP server is a connector that exposes parts of the Google Ads API as tools an AI assistant can call. MCP, the Model Context Protocol, is an open standard in which an AI application (the host) talks to servers that offer tools, resources and prompts. Instead of exporting a report and pasting it into a chat, you let the assistant ask the account directly.
The server does not add intelligence to your account. It turns a request such as "list campaigns with the highest spend" into a query against the Google Ads API and returns the rows. The assistant then reads, compares and explains them. Our guide to Claude and MCP covers the client side.
Is there an official Google Ads MCP server?
Yes. Google publishes an open-source Google Ads MCP server, and the Google Ads API developer site documents it under its developer toolkit.
Here is what the sources say, as of October 2026:
- The code lives at
github.com/googleads/google-ads-mcp, in thegoogleadsorganization, under the Apache-2.0 licence. The repository was created in October 2025, and its latest release at the time of writing is version 0.0.5, published on 9 October 2026. - Google's documentation page (last updated 16 October 2026) describes it as a bridge between AI agents and the Google Ads API, written in Python, and marks the current release as read-only.
- The documentation lists three tools:
list_accessible_customers,searchandget_resource_metadata. - The repository also offers four resources that help a model write correct queries: the API discovery document, the available metrics, the available segments and the release notes.
The version number matters. A 0.0.x release is early software, so expect names, options and setup steps to change. Pin the version you tested and read the release notes before upgrading.
There are also community-built servers, some with write tools. We have not reviewed them here and do not name any. If you consider one, treat it like any third-party code that will hold credentials for your ad account: read the source, check the tool list for anything that edits, and prefer the official server unless you have a clear reason.
What can an assistant do with it?
With the official server, an assistant can read account data and reason over it. The search tool runs queries written in the Google Ads Query Language (GAQL), and the other two tools list the accounts the user can reach and describe which fields each resource offers.
In practice, that covers the diagnostic work that usually eats an hour of exports:
- Campaign and budget status. Which campaigns are active, limited by budget or paused, and how they are set up.
- Performance by segment. Spend, clicks and conversions broken down by campaign, ad group, device or day.
- Search terms. The queries that triggered your ads, which is where wasted spend and missed ideas show up.
- Ad and asset review. The text of ads and assets in use, so the assistant can compare them with your landing pages.
- Cross-checks. Comparing two periods, or two campaigns with similar goals.
What it cannot do is as important. Google's documentation states that the read-only release cannot modify bids, pause campaigns or create new assets, and that sending mutate operations is outside the server's scope. It does not replace judgement either: an assistant can spot that a campaign spent more this week, but it does not know that you ran a promotion or paused a product line unless you tell it.
What changes for a marketer?
The change is speed of access, not a new kind of analysis. Questions that used to need a saved report, a spreadsheet and some patience become a sentence, and the assistant can chain several queries to answer a follow-up.
Routine reviews get shorter, because the assistant does the pulling and sorting and you do the deciding. More people can look at the account without learning the interface. And the quality of the question matters more: a confident summary of the wrong rows is worse than no summary.
The numbers still come from Google Ads, and the assistant is a reader on top of them. For anything that drives a decision, open the account and check the figure it quoted. If you manage paid campaigns as part of a broader plan, this fits inside your digital marketing strategy as a faster way to read data, not as a replacement for the people who set goals and budgets. For campaign work itself, see how we approach Google Ads management.
Want a safe review of your Google Ads account?
Tell us what you run and what you want to understand. We reply with how we would set up the reading and the checks.
See our Google Ads workHow do you connect an AI assistant to Google Ads?
You connect it by running the server, giving it credentials for a Google Ads user, and adding it to your assistant's MCP configuration. The steps below follow the repository's README and Google's documentation as of October 2026. Check both before you start, because the project is moving.
- Pick the right account structure. Decide which Google Ads account, or manager account, the assistant should reach. If you reach client accounts through a manager account, you will need the manager's customer ID.
- Enable the API. In a Google Cloud project, enable the Google Ads API. According to Google, enabling it gives the project Test access, which is meant for test accounts.
- Get production access. To read real accounts, the Google Cloud project needs a higher access level. Google's page lists Explorer, Basic and Standard, applied for in the Google Ads API Overview page of the Cloud Console. The README says the access must be at least Explorer to query production accounts.
- Choose how to authenticate. The README offers an OAuth proxy for a server run as a web service, Application Default Credentials with the
https://www.googleapis.com/auth/adwordsscope, or agoogle-ads.yamlfile used by the Google Ads Python client library. Google's page also mentions OAuth 2.0 and service accounts. - Set the login customer ID if you use a manager account. Google's call-structure documentation says the
login-customer-idheader must be the manager account's customer ID, written without hyphens. Leaving it out when you authenticate through a manager account causes aUSER_PERMISSION_DENIEDerror. - Run the server and add it to your client. The README shows running it with
pipx, either from the repository or from a pinned PyPI version, and gives a JSON configuration for MCP clients that start local servers. Copy the current snippet from the README rather than from a blog post, since variable names change between releases. - Test with a harmless question. Ask it to list the accounts it can reach, then to describe the campaign resource, before asking anything about spend.
A note on the developer token. For years, every Google Ads API call needed one. Google's documentation says developer tokens were sunset on 9 September 2026: sending one is now optional and the API ignores it, and access levels are granted to the Google Cloud project instead. The server's README still has a "Configure Developer Token (Optional)" step, so older tutorials that treat the token as mandatory may be out of date. For the exact environment variables, follow the README of the version you install.
Two warnings from the documentation are worth repeating. The README states that the server "will expose your data to the Agent or LLM that you connect to it." And Google's page warns that deploying the server to Cloud Run with unauthenticated access, without the OAuth client ID and secret, could expose the container's service account credentials to unauthenticated callers.
Which safeguards should you set first?
Set the smallest possible access, keep it read-only, require a person to approve any change, and keep a record of what the assistant did. The order matters: the first two controls stop most problems before they begin.
Start read-only. The official server is read-only today, which is the right place to begin. Keep it that way until you have used it for weeks and know where it makes mistakes. If a server offers write tools, switch them off in its configuration where possible. The README describes a tools_config.yaml file that lets you enable or disable individual tools, which is a practical way to expose only search and the listing tools.
Match the account user to the job. Google Ads gives each user an access level: email-only, read-only, billing, standard or admin. Google's help page describes read-only as able to view campaigns and use planning tools, standard as able to edit campaigns, and admin as able to manage access and links. Authenticate the assistant as a user with read-only access, so that even a faulty tool cannot edit. This is separate from the API access level, which governs usage limits, not what a user may change.
Use a separate credential. Do not reuse the credentials of the person who owns the account. Create a dedicated Google user or service account, grant it the minimum level on the accounts it needs, and keep the secrets out of the MCP client's configuration file where you can. The README itself advises keeping server credentials out of the client configuration for the HTTP setup.
Put a human in front of every change. The MCP specification says hosts must obtain explicit user consent before invoking any tool, and that users must keep control over what actions are taken. Apply the same idea to your process: if the assistant suggests a budget or bid change, a person reads the proposal, opens Google Ads, and makes the change there. No automation should adjust spend without a review.
Log and review. Keep a record of which questions were asked and which queries ran. Review it after the first weeks, and revoke access when someone leaves the project. You can revoke the credential, remove the user from the account or delete the OAuth grant in your Google account settings.
Treat the model as a data recipient. Whatever the server returns is sent to the model provider you chose. Check that provider's data terms before you connect.
What should you do about changes to budgets and bids?
Leave them to a person. Budgets, bids, targeting, ads and conversion settings change what you spend and what you measure, so a mistake costs money immediately and can be hard to spot afterwards.
Look at two things: whether the action reads or changes the account, and how much money rides on it. Reading is low risk. Changing is where the controls belong, and the more spend an action touches, the more review it needs.
If you later add a tool that can write, apply the same grid: start with low-stakes changes and require a person to confirm each one. Because the official server does not send mutate operations, a write step would involve another tool or a person acting in the interface. Google's documentation points to a separate Developer Assistant for generating code that writes, which is a developer workflow with its own review, not a switch you flip in the chat.
Tracking deserves the same caution. Conversion actions decide what Google's bidding learns from, so a careless change can quietly distort every campaign. For the measurement side, see how we think about analytics and tracking, and the sibling guide on connecting an assistant to Google Analytics.
Need an assistant that reads your data and stops at the edit button?
We design AI workflows with read-only access, human approval and logs. Tell us which reports eat your week.
See AI workflowsWhich questions are worth asking?
The best questions are specific, tied to a period and a goal, and end with something you can check. Here are seven that work well with read-only access. Adapt the wording to your account, and always say the date range.
- Wasted search terms. "List search terms from the last 30 days that spent money and produced no conversions. Group them by theme and suggest which could become negative keywords."
- Spend gaps. "Compare spend by campaign this week with the previous week. Which campaigns moved the most, and what changed in their settings that could explain it?"
- Budget-limited campaigns. "Which campaigns are limited by budget while their conversions look healthy, and which are not limited but barely spend?"
- Ad quality. "Show the ads in my best campaign. Compare the headlines with the landing page promise and point out mismatches."
- Device and time splits. "Does performance differ by device or by day of week for this campaign? Show the rows behind each claim."
- Overlap and cannibalisation. "Do any two campaigns bid on very similar search terms? List the terms and the campaigns involved."
- Account hygiene. "List campaigns with no activity in the last 60 days and ad groups with no active ads, so I can clean up."
Ask the assistant to show the query it ran and the rows it used, and ask what evidence would prove its recommendation wrong. Then verify the key figures in Google Ads before you act.
Avoid asking it to "optimise" without a goal or to decide how much to spend. Those decisions belong to people who know the margins and the plan.
Should you use the Google Ads API directly instead?
Use the API directly when you are building a product or an automated pipeline, and use an MCP server when you want an assistant to explore the account in conversation. They are the same data with different front doors.
The API is what developers call from code. It suits dashboards, nightly exports and anything that must run the same way every time. An MCP server suits open-ended questions where you do not know in advance which query you need.
For repeating work, such as a weekly report that goes to a client, a fixed pipeline is usually more reliable than a conversation, because the same query produces the same table each time. You can combine them: use the assistant to explore and find a useful query, then turn it into a scheduled report that a person reviews. That is the kind of job we scope as AI workflows. If you want to compare this with other connectors, our overview of MCP tools for marketers lays out the options.
Frequently Asked Questions
What is a Google Ads MCP server?
A Google Ads MCP server is a program that gives an AI assistant tools to query a Google Ads account through the Google Ads API. It follows the Model Context Protocol, so any compatible assistant can use it. The official server runs the queries and returns the results to the assistant, which explains them in plain language.
Is the Google Ads MCP server official?
Yes. Google publishes an open-source server in the googleads GitHub organization and documents it on the Google Ads API developer site. As of October 2026, its latest release is 0.0.5. Community servers also exist, but they are not from Google, and you should review their code before use.
Can an AI assistant change my Google Ads campaigns through MCP?
Not with the official server today. Google's documentation says the current release is read-only and cannot modify bids, pause campaigns or create assets. If you use another server that can write, restrict its tools, use a limited user and require a person to approve every change.
Do I still need a developer token?
Google's documentation says developer tokens were sunset on 9 September 2026 and that sending one is optional and ignored. Access levels now belong to your Google Cloud project. The server's README still lists the token as an optional step, so follow the instructions for the version you install.
What access does the assistant need?
It needs a Google Ads user or service account that can reach the accounts you want to read, plus a Google Cloud project with the Google Ads API enabled and a production access level such as Explorer. For safety, give the user read-only access in Google Ads so that it cannot edit anything.
Is it safe to connect my ad account to an AI model?
It can be safe if you limit it: read-only access, a dedicated credential, human approval of every change, and logs. The data you query is sent to the model provider, so check their terms first.
Do I need to know how to code to use it?
The setup is technical: enabling the API, creating credentials and editing a configuration file. After that, you ask questions in plain English.
Next steps
Start small: read-only access, one account, and a handful of questions you already know the answer to, so you can judge the quality. Then widen the scope slowly and keep a person in front of every change. If you want help with campaign work, see our Google Ads work and the other services linked above, or get in touch with the account you manage and the questions you want answered.
Sources
- Google Ads API, "Google Ads API MCP server" (developer toolkit, last updated 16 October 2026): https://developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server
- GitHub, googleads/google-ads-mcp (README, tools, setup, releases, Apache-2.0 licence): https://github.com/googleads/google-ads-mcp
- Google Ads API, "Obtaining a developer token" (sunset on 9 September 2026, access levels by Cloud project): https://developers.google.com/google-ads/api/docs/get-started/dev-token
- Google Ads API, "Call structure" (
login-customer-idfor manager accounts): https://developers.google.com/google-ads/api/docs/concepts/call-structure - Google Ads API, "OAuth 2.0 overview": https://developers.google.com/google-ads/api/docs/oauth/overview
- Google Ads Help, "About access levels in your account": https://support.google.com/google-ads/answer/9978556
- Model Context Protocol, Specification 2025-11-25, "Security and Trust & Safety": https://modelcontextprotocol.io/specification/2025-11-25/index
Newsletter
One short email when we publish
Notes on AI, image, video and brand work, from the people who do it. No spam.