What the Google Ads MCP server does
MCP (Model Context Protocol) is an open standard for connecting AI assistants to outside tools and data. Google publishes an official Google Ads MCP server (source on GitHub) that turns the Google Ads API into tools an assistant can call.
Once it is connected, you can ask things like:
- "Which campaigns spent the most last week, and what was their ROAS?"
- "List search terms with more than 50 clicks and no conversions."
- "Which ad groups have no sitelinks?"
The assistant writes and runs the query, then explains the result.
Before you start
You need:
- Python 3.10 or newer and pipx.
- The gcloud CLI (Google Cloud SDK).
- A Google Cloud project with the Google Ads API enabled and an active billing account. Projects on a free trial or with a suspended billing account are rejected.
- API access approved for that project. Since September 9, 2026, Google no longer uses developer tokens. Access levels (Test, Explorer, Basic, Standard) belong to the Cloud project that owns your OAuth client. Apply from the Google Ads API page in the Cloud Console; new Basic and Standard applications require brand verification first. A Test-level project cannot reach live accounts and returns
CLOUD_PROJECT_NOT_APPROVED_FOR_PRODUCTION.
Step-by-step setup (local, with Claude or Gemini)
1. Sign in with application default credentials
Run this and complete the browser sign-in with the Google account that can access your ads account:
gcloud auth application-default login --scopes=https://www.googleapis.com/auth/adwords,https://www.googleapis.com/auth/cloud-platform
gcloud prints the path of the credentials file it saved. Copy it for the next step.
2. Add the server to your assistant's settings
Open ~/.claude/settings.json (Claude) or ~/.gemini/settings.json (Gemini CLI) and add:
{
"mcpServers": {
"google-ads-mcp": {
"command": "pipx",
"args": ["run", "--spec", "google-ads-mcp==0.0.6", "google-ads-mcp"],
"env": {
"GOOGLE_CLOUD_PROJECT": "PROJECT_ID",
"GOOGLE_APPLICATION_CREDENTIALS": "PATH_TO_CREDENTIALS_JSON"
}
}
}
}
Replace PROJECT_ID with your Google Cloud project ID and PATH_TO_CREDENTIALS_JSON with the path from step 1. Version 0.0.6 is current at the time of writing; check the GitHub page for newer releases.
3. Add your manager account (agencies only)
If you reach client accounts through a manager (MCC) account, add "GOOGLE_ADS_LOGIN_CUSTOMER_ID": "1234567890" to the env block, using the 10-digit manager ID without dashes.
4. Test it
Restart the assistant and ask: "List the Google Ads accounts I can access." It should call customers_list_accessible_customers and return your customer IDs.
Remote setup
The server can also run over streamable HTTP, for example on Cloud Run, with an OAuth web client instead of local credentials. That lets a team share one server, but you have to deploy, secure and maintain it yourself.
What the official server can and cannot do
| Task | Official server |
|---|---|
| List accessible accounts | Yes (customers_list_accessible_customers) |
| Run reports with GAQL queries | Yes (search_search) |
| Look up fields and resources | Yes (metadata_get_resource_metadata) |
| Create text, image, video, sitelink, callout and structured snippet assets | Yes (assets_create_asset) |
| Upload a video asset | Local mode only |
| Create and attach labels | Yes |
| Create campaigns or ad groups | No |
| Change budgets or bids | No |
| Create responsive search ads | No |
| Add keywords or negative keywords | No |
Writes are proposed before they happen: the server first validates the change without applying it, returns a checksum, and only commits after you confirm. You can also run it in read-only mode.
In practice the official server is excellent for analysis and light asset work, but you still build and edit campaigns in the Google Ads interface.
Hosted alternatives
Several companies now offer hosted Google Ads MCP servers. They usually trade the local setup for a one-click OAuth sign-in, and some add more write actions. Before connecting one, check:
- Which write actions it allows, and whether changes are applied as drafts or paused.
- Whether every change needs your confirmation, and whether there is a log of what the assistant did.
- Where your tokens are stored and whether you can revoke access in one place.
- Spending guardrails, such as limits on budget changes.
Forespend MCP (in development)
Forespend is building a hosted Google Ads MCP server with profit guardrails: connect with Google sign-in, no Python or gcloud, and let your assistant create campaigns and responsive search ads as paused drafts that you review before anything spends. It is not available yet. Contact us to join the early-access list.
While you wait, two free tools help you sanity-check what any assistant suggests: the ROAS calculator and the Google Ads cost calculator.