Google Open-Sources an MCP Server for Google Ads — LLMs Can Query Ads Data Read-Only
What Google released
Google published an open-source Model Context Protocol (MCP) server that provides read-only access to the Google Ads API for agentic and LLM applications. The repository googleads/google-ads-mcp implements the server in Python and currently exposes two tools: search (GAQL queries over Ads accounts) and list_accessible_customers (enumeration of customer resources). The project is distributed under the Apache-2.0 license and marked as “Experimental” in the repository.
MCP tooling exposed
The MCP server advertises small, typed “tools” that models and agent frameworks can discover and call. For Google Ads the server maps those tools to Google Ads API operations, allowing LLM agents to request campaign telemetry, performance metrics, and account listings through natural-language-driven workflows without writing bespoke SDK glue.
How it works: protocol, discovery, and invocation
MCP standardizes how tools are described and invoked: clients discover available tools and their schemas, then call them with typed parameters and receive typed responses. MCP-compatible clients, such as Gemini CLI and Code Assist, can point to this Ads MCP server and run GAQL queries or enumerate accessible customers during a session.
Authentication and setup
To use the server you enable the Google Ads API in a Cloud project, obtain a developer token, and configure OAuth2 credentials. The Ads MCP server requires the https://www.googleapis.com/auth/adwords scope. For manager-account hierarchies you can configure a login customer ID so the server queries the correct account context.
Client wiring and invocation
The repository includes guidance for running the MCP server and integrating it with MCP clients. A common setup is to run the package via pipx and then configure a MCP client (for example, Gemini CLI) to point at the server, passing credentials via environment variables. The project mentions the following invocation as an example of how to run the server:
pipx run git+https://github.com/googleads/google-ads-mcp.git google-ads-mcp
Once configured, clients can query the server via an MCP endpoint (for example, /mcp in Gemini) and prompt agents to run GAQL search queries or call list_accessible_customers to discover customer IDs.
Why this matters
MCP is emerging as a practical interoperability layer to wire large models to external systems. By offering a reference Ads MCP server, Google lowers the integration cost for teams experimenting with LLM-driven workflows—such as anomaly triage, weekly reporting, or QA on campaign performance—without granting write permissions. This read-only reference lets teams validate agent-assisted analysis and reporting while avoiding production write risks.
Limitations and cautions
The project is flagged Experimental, and the server exposes only read-only capabilities today. Production use requires careful management of developer tokens, OAuth scopes, and account access. The README and repository pages note data-exposure considerations and best practices for credentials.
Ecosystem signal
The release arrives as MCP adoption expands across vendors and open-source clients, reinforcing MCP as a pragmatic path for agent-to-SaaS interoperability. For PPC and growth teams, the reference server is a low-friction way to test LLM-native access patterns to Ads data before investing in custom integrations.