What is the Google Ads MCP?

MCP (Model Context Protocol) is an open standard that allows AI models to connect with external tools securely. Google released its own MCP server for Google Ads, which means tools like Claude can query your campaign data directly from the API, without opening the Google Ads interface.

The server is open source, developed by the official Google Ads team, and operates in read-only mode — it cannot modify campaigns, budgets or bids. It only reads data.

What can you do with Claude + Google Ads?

With this integration you can:

Audit conversion configuration — verify which are primary vs secondary, if counting is correct (ONE_PER_CLICK for leads), and if values are properly assigned

Analyze metrics in natural language — ask "what's my CPA for the last 30 days?" and get the answer with context

Detect budget-burning keywords — find keywords with high spend and zero conversions in seconds

Review campaign structure — validate that ad groups make sense, extensions are active, and ad schedules are correct

Compare performance across periods — "did my ROAS improve this month vs last?" with real data

Identify undersized audiences — detect remarketing lists with fewer than 1,000 users that can't segment

How to set it up

Requirements:

Google Ads account with read access

Google Ads developer token (obtained from the developer console)

OAuth 2.0 credentials (client ID and client secret from a Google Cloud project)

Python installed (for the MCP server)

Step by step:

1. Install the MCP server — install via pip or pipx from Google's official package

2. Configure credentials — add them as environment variables (developer token, client ID, client secret)

3. Connect with Claude — add the server to Claude Code or Claude Desktop MCP configuration

4. Verify access — use "list my accessible accounts" to confirm the connection

3 available tools:

list_accessible_customers — lists all Google Ads accounts you have access to

search — runs GAQL (Google Ads Query Language) queries to fetch any metric or account data

get_resource_metadata — returns API resource structure to understand what fields you can query

3 prompts you can copy and use

Prompt 1 — Conversion audit:

"Show me all conversions for account [ID]. For each one tell me: name, whether it's primary or secondary, counting type (ONE_PER_CLICK or MANY_PER_CLICK), conversion window, and assigned value. Flag any with problematic configuration for lead generation."

Prompt 2 — Underperforming keywords:

"Analyze keywords from the last 30 days for account [ID]. List all with more than $50 spend and zero conversions. Sort by spend descending and include: keyword, match type, ad group, campaign, impressions, clicks, cost and CTR."

Prompt 3 — Performance comparison:

"Compare performance of all active campaigns in account [ID] between the last 30 days and the previous 30 days. For each campaign show: name, spend, conversions, CPA and ROAS for both periods, and the percentage change."

Limitations

Read-only — cannot pause campaigns, change bids or modify ads. The current MCP server version is read-only

Requires developer token — you need Google Ads API access, a regular advertiser account isn't enough

Uses GAQL — internal queries use Google Ads Query Language. Claude handles it well, but very complex queries may require iteration

Doesn't replace human auditing — Claude detects patterns and numbers, but strategic interpretation requires business context

Official source

This guide is based on the official Google Ads MCP server documentation published by Google. You can find the full documentation, source code and installation instructions on the official Google Developers site.

Source: developers.google.com/google-ads/api/docs/developer-toolkit/mcp-server

Claude + Google Ads: How to Use AI to Audit and Optimize Your Campaigns

Natural language auditing

Read-only — zero risk

Metrics in seconds

Error detection