Google Ads MCP: How to Connect Google Ads to ChatGPT and Claude in 2026

What is Google Ads MCP?

Google Ads MCP is a Model Context Protocol server that translates an AI assistant's tool calls into requests to the Google Ads API. Instead of exporting reports, pasting spreadsheets, or teaching the model your account structure in every chat, the MCP server gives an authorized assistant a defined set of tools for retrieving live account information.

The important distinction is that the model does not become Google Ads. The AI client plans the request, the MCP server enforces the available tool contract, and Google Ads remains the system of record. Authentication, account access, API limits, and manager-account rules still apply.

What can the official Google Ads MCP server do?

The official server is most useful when you want an assistant to answer account questions from current Google Ads data. It can run reporting queries, inspect available fields, discover directly accessible customer IDs, and expose reference resources describing metrics, segments, API resources, and release notes.

• Compare campaign spend, conversions, CPA, ROAS, clicks, and impression share for a defined date range.
• Find campaigns or ad groups whose performance changed week over week.
• Inspect search-term and keyword data available through Google Ads Query Language.
• List customer accounts that the authenticated Google user can access directly.
• Ask the model which resource fields and segments are available before it writes a query.

It is not currently a complete campaign-management agent. If a connector advertises budget changes, campaign creation, pausing, keyword edits, or other mutations, that functionality comes from a different or extended MCP server. Verify who operates that server, which Google OAuth scopes it requests, and where its tokens are stored before connecting it to a production account.

Choose the right connection model

Claude Desktop and developer tools can run local MCP processes. A browser-based AI product cannot reach a process running only on your laptop, so it normally needs a remote MCP server exposed over HTTPS. Google's repository documents a Cloud Run deployment using the FastMCP OAuth proxy for that pattern.

For nontechnical advertisers, a managed connection removes the server deployment, credential-file editing, updates, and uptime work. That convenience is useful only if the provider offers transparent permissions, token encryption, revocation, audit logs, tenant isolation, and conservative approval controls.

What you need before connecting Google Ads

• A Google user with access to the Google Ads customer account or its manager account.
• A Google Cloud project with the Google Ads API enabled.
• A Google Ads developer token with an access level suitable for the accounts you want to query.
• OAuth 2.0 credentials or Application Default Credentials that include the Google Ads scope.
• The ten-digit Google Ads customer ID, entered without dashes when a configuration field requires it.
• For manager-account access, the login customer ID of the manager account through which access is granted.

How to set up Google's official MCP server

Google's maintained README is the source of truth for commands and environment variables. The sequence below explains the architecture so you know what each setup screen is accomplishing; confirm command syntax against the repository before deploying because the project is actively maintained.

1. 1

   Create or select a Google Cloud project

   Use a project owned by your business, verify billing and administrator access, then enable the Google Ads API. Avoid placing a client production integration inside a personal or disposable project.

2. 2

   Obtain the Google Ads developer token

   Open API Center from a Google Ads manager account. The token's access level determines whether it can query test or production accounts. Keep the token server-side and never paste it into an article, screenshot, browser chat, or public repository.

3. 3

   Create OAuth credentials

   Configure an OAuth consent screen and client credentials. The official server supports a FastMCP OAuth proxy for remote user authorization, Application Default Credentials, or an existing Google Ads API client configuration.

4. 4

   Install and run the server

   For a local client, the repository documents running the package with pipx from the official GitHub source. For a remote client, deploy the container to Cloud Run or equivalent infrastructure and expose the streamable HTTP endpoint over HTTPS.

5. 5

   Add the MCP server to your AI client

   Local clients use a command-based server configuration. Remote-capable clients use the HTTPS MCP endpoint and complete the OAuth flow. Enable only the tools needed for the task.

6. 6

   Test with a read-only account question

   Start by listing accessible customers, then query a narrow date range for one known customer ID. Confirm that spend and conversion totals match Google Ads before trusting broader analysis.

How the ChatGPT connection works

ChatGPT connects to remote MCP servers rather than launching your local Python process. That means the Google Ads MCP server must be available through a secure remote endpoint and implement the authentication flow expected by the client. In managed workspaces, an administrator may need to review, publish, or allow the custom MCP app before members can use it.

OpenAI's current documentation describes full MCP support, including write actions, as a beta capability for certain ChatGPT workspace plans. Availability, interface labels, and administrator controls can change. Check the current plan documentation before promising a one-click connection to every ChatGPT user.

How the Claude connection works

Claude supports both local and remote MCP patterns, but they are configured differently. Local servers are appropriate for Claude Desktop or Claude Code on a machine that can launch the process. Remote connectors are added in Claude's connector settings and are available on supported paid plans.

1. 1

   For a remote server

   Open Settings, go to Connectors, add the trusted remote MCP URL, and complete OAuth. Then enable the connector from the Search and tools menu in the conversation.

2. 2

   For a local server

   Use the current Claude Desktop extension or Claude Code MCP setup flow. Keep credentials in environment variables or an approved local credential store, not inside prompts.

3. 3

   Review each tool request

   For unfamiliar connectors, avoid permanent approval. Read the tool name and inputs, especially if a third-party server adds mutation tools beyond Google's official read-oriented surface.

Security checklist before you authorize an ads account

• Confirm the exact MCP URL from the provider's own domain; do not trust a connector URL copied from an anonymous post.
• Review the Google account, customer accounts, and OAuth scopes shown on the consent screen.
• Keep developer tokens, client secrets, refresh tokens, and signing keys outside browser-delivered JavaScript.
• Encrypt OAuth tokens at rest and separate each customer's credentials and account mappings.
• Log tool calls, the initiating user, customer ID, inputs, outputs, and approval state without logging secrets.
• Use read-only tools for reporting workflows and place material budget, status, targeting, and deletion actions behind explicit approval.
• Provide a visible disconnect flow and document how users can revoke access from their Google account.

MCP is a protocol, not a security certification. A server can expose only safe reporting tools or it can expose destructive write tools. The operator, hosting design, tool definitions, OAuth scopes, and approval policy determine the actual risk.

Common Google Ads MCP setup problems

When debugging, reduce the request to one customer, one date range, and one metric. Verify authentication and raw totals before asking the model for explanations. A fluent narrative is not evidence that the underlying query used the correct account or attribution settings.

Five useful prompts after the connection works

• List the Google Ads customer IDs I can access and identify which are manager accounts.
• For customer ID [ID], compare spend, conversions, conversion value, CPA, and ROAS for the last 7 days versus the previous 7 days. State the exact date ranges used.
• Show campaigns with more than 20% week-over-week spend growth and declining conversion volume. Do not recommend changes until the data table is shown.
• Find search terms with meaningful spend and zero conversions in the last 30 days. Include campaign, ad group, match type, spend, clicks, and conversion value.
• Explain which Google Ads fields and segments are available for diagnosing a drop in impression share, then propose the query before running it.

Good prompts define the customer ID, comparison period, business metric, minimum-data threshold, and expected output. They also ask the assistant to show evidence before interpreting it. That discipline matters more than adding generic instructions such as 'act like an expert media buyer.'

The bottom line

Google Ads MCP can replace repetitive report exports and let ChatGPT or Claude reason over current account data, but the connection is only as reliable as its authentication, account routing, queries, and security controls. Google's official server is the safest reference point for read-oriented access. A managed connector can make onboarding much easier, particularly for marketers who do not want to run Python or Cloud Run, but it should earn trust with clear permissions and auditable behavior.

For teams building a production SaaS around this workflow, the real product is not simply an MCP URL. It is reliable OAuth onboarding, tenant isolation, account selection, secure token storage, plan-aware instructions for each AI client, monitoring, rate-limit handling, approval-gated actions, and support when Google changes its API or authentication requirements.

Frequently asked questions