← MCP directory

Google Ads MCP server

official

Google publishes its own MCP server for the Google Ads API that lets AI agents fetch account and campaign data through GAQL queries, create assets and manage labels. The server runs locally or on Cloud Run and can be set to read-only.

What an agent can do

  • → List accessible ad accounts
  • → Fetch campaign performance via GAQL
  • → Analyse keywords, search terms and budgets
  • → Create assets and upload video
  • → Create labels and apply them to campaigns

The server cannot change bids, pause campaigns or create ads, and it cannot link assets to campaigns. Since version 0.0.6 it creates assets and labels by default, after a propose-and-confirm step, and that is switched off in tools_config.yaml.

About the Google Ads MCP server

The Google Ads MCP server is an open source program that Google publishes in the googleads organisation on GitHub and as the package google-ads-mcp. It is a bridge between an AI assistant and the Google Ads API. Google does not host a shared address for it. You run it yourselves, either locally on a computer or as your own service on Google Cloud Run that several people can connect to.

Teams use it to ask about the account in their own words: cost and conversions per campaign, search terms that eat budget, budgets and campaign status. The server builds the Google Ads query for the assistant and can look up which fields are selectable, filterable and sortable. Since version 0.0.6 it can also create assets such as texts, images, sitelinks and YouTube videos, upload video and manage labels.

The server cannot create, change, pause or enable campaigns, ad groups, ads, budgets or bids, and it cannot link assets to campaigns. The changes it can make go in two steps: first a proposal that is validated, then a confirmation with a checksum. For strictly read-only use, switch off the label and asset categories in the configuration file. Access to real accounts is decided by the access level of the Google Cloud project.

Things to ask

  • Which accounts can I access, and which one is our main account?
  • Show cost, clicks and conversions per campaign for the last 30 days, sorted by cost.
  • Which search terms cost the most this month without a single conversion?
  • Which campaigns are limited by their budget right now?
  • Create a label called Q4 test and propose which campaigns it should be put on.

Tools the server exposes

  • customers_list_accessible_customers
  • search_search
  • metadata_get_resource_metadata
  • assets_create_asset
  • assets_upload_video_asset
  • labels_manage_labels
  • labels_manage_label_associations
GuideHow to set up Google Ads MCP →

Part of these packages

Google Ads is part of these Walma packages.

Questions about Google Ads MCP

Is the Google Ads MCP server free?+

The server is open source under the Apache 2.0 licence and is available on GitHub and as the package google-ads-mcp. Google states no charge for the server itself. You need a Google Cloud project, and that project's access level to the Google Ads API decides whether it reaches real accounts. If you run it on Cloud Run, that happens in your own project.

What is the Google Ads MCP server URL?+

There is no shared address at Google. The server is the package google-ads-mcp, with the source at github.com/googleads/google-ads-mcp. Locally it is started with pipx. If you deploy it to Google Cloud Run you get your own service URL, and clients connect to that URL with /mcp at the end.

How does sign-in work?+

Locally the server uses Application Default Credentials, either from an OAuth client file with your own Google sign-in or from a service account. As a remote service on Cloud Run it uses an OAuth client of the web application type, and every tool call then runs under the signed-in person's identity. The scope it needs is adwords.

Is it read-only?+

Not by default. Besides the read tools there are tools that create assets, upload video and manage labels. For strictly read-only use, set labels and assets to false in tools_config.yaml. The server can never create, change, pause or enable campaigns, ad groups, ads, budgets or bids.

Which AI assistants does it work with?+

Google's documentation shows Gemini CLI, Google Antigravity and Claude, and names Claude and ChatGPT as remote connectors to a Cloud Run deployment. The project README also names Claude Code, Cursor, VS Code with Copilot and Codex CLI. Clients without OAuth support get a 401 from the remote server and have to use the local setup.

How is a change confirmed before it is made?+

Every write tool starts with a proposal that is validated without changing anything. The proposal returns a list of changes and a checksum. The change is only carried out when the tool is called again with the same arguments and the matching checksum. Google points out that the checksum is not access control: to block changes entirely, switch off the categories.

Why does the server only reach test accounts?+

The access level is set by the Google Cloud project that owns the OAuth client or the service account. New projects get Test access, which covers test accounts only. Real accounts need Explorer, Basic or Standard access. Without it, calls fail with the error CLOUD_PROJECT_NOT_APPROVED_FOR_PRODUCTION.

Missing a connection?

We build and review MCP servers for teams that need them, and run them behind the same policy and log as the rest of the gateway.