google-ads-mcp-server
An MCP server that provides read and write access to Google Ads, allowing natural language management of campaigns, budgets, ad groups, bids, and keywords, with dry-run validation for safety.
README
google-ads-mcp-server
A Model Context Protocol (MCP) server for Google Ads with both read and write access. Connect it to Claude (or any MCP client) and pull reports with GAQL, then actually manage the account — create and adjust budgets, campaigns, ad groups, bids, and keywords — in natural language, using your own Google Ads credentials.
Google's own official Google Ads MCP server is read-only. This one adds the write side, and does it carefully.
🛑 Writes spend real money — how the safety guard works
Google Ads changes cost money, so every write tool is a dry run by default. When you call a write tool without confirm=True, the change is sent to the Google Ads API with its validate_only flag: the API checks that the change is valid but does not apply it, and the tool tells you it was validated only. The change is applied only when you call the same tool again with confirm=True.
That means the safe loop is always: call once to validate → review → call again with confirm=True to apply. New campaigns are also created PAUSED by default so nothing starts serving unexpectedly.
Features
Read
ads_whoamiandlist_accessible_customers— verify credentials and see which accounts you can reachsearch— run any GAQL query (the general-purpose read tool)- Convenience listers: campaigns, ad groups, ads, budgets, keywords
campaign_performance— impressions/clicks/cost/conversions by campaign over a date range
Write (dry run unless confirm=True)
- Create and update campaign budgets
- Create campaigns (paused Search + Manual CPC by default) and set campaign status (enable/pause/remove)
- Create ad groups, set ad group status, and change ad group bids
- Add keywords and set keyword status
Requirements
- Python 3.10+
- A Google Ads developer token. Test tokens only work against test accounts; to read or write real accounts you need Basic access, which is a short application in the API Center of a manager (MCC) account.
- Credentials — either OAuth2 (client id, client secret, refresh token) or a service account with domain-wide delegation (Google Workspace).
- Your manager account id as the login customer id.
Google setup (one time)
- In a Google Ads manager (MCC) account, open API Center and apply for a developer token with Basic access.
- In the Google Cloud Console, create a project and enable the Google Ads API.
- Create OAuth 2.0 credentials (Desktop app) and generate a refresh token for an account that can access your Google Ads accounts. (Google's OAuth desktop guide walks through this; the
google-adslibrary ships a helper script for the refresh token.) - Note your login customer id (the MCC id, digits only).
Alternatively, use a service account with domain-wide delegation on a Workspace and point the config at its key — see the google-ads configuration docs.
Install
Using uv (recommended):
git clone https://github.com/burhan29ee/google-ads-mcp-server.git
cd google-ads-mcp-server
uv venv --python 3.12 .venv
uv pip install --python .venv/bin/python -e .
Or with pip:
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
Configure your MCP client
The server reads the standard GOOGLE_ADS_* environment variables (or a google-ads.yaml via GOOGLE_ADS_CONFIGURATION_FILE_PATH). For Claude Desktop, edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json; Windows: %APPDATA%\Claude\claude_desktop_config.json):
{
"mcpServers": {
"google-ads": {
"command": "/absolute/path/to/google-ads-mcp-server/.venv/bin/python",
"args": ["-m", "google_ads_mcp_server.server"],
"env": {
"GOOGLE_ADS_DEVELOPER_TOKEN": "your-developer-token",
"GOOGLE_ADS_LOGIN_CUSTOMER_ID": "1234567890",
"GOOGLE_ADS_CLIENT_ID": "your-oauth-client-id.apps.googleusercontent.com",
"GOOGLE_ADS_CLIENT_SECRET": "your-oauth-client-secret",
"GOOGLE_ADS_REFRESH_TOKEN": "your-oauth-refresh-token"
}
}
}
}
Restart the client and start with "run ads_whoami" to confirm access. Then try "show my top campaigns by spend over the last 30 days" or "pause campaign 12345" (which will dry-run first).
Tool reference
| Tool | Type | What it does |
|---|---|---|
ads_whoami |
read | Show config (masked token, login id) and accessible accounts |
list_accessible_customers |
read | Customer resource names this login can access |
search |
read | Run any GAQL query |
list_campaigns |
read | Campaigns with status, channel, budget |
list_ad_groups |
read | Ad groups (optionally by campaign) |
list_ads |
read | Ads (optionally by ad group) |
list_campaign_budgets |
read | Budgets with amount and delivery |
list_keywords |
read | Keyword criteria (optionally by ad group) |
campaign_performance |
read | Per-campaign metrics over a date range |
create_campaign_budget |
write | Create a budget (dry run unless confirm) |
update_campaign_budget |
write | Change a budget amount (dry run unless confirm) |
create_campaign |
write | Create a paused Search campaign (dry run unless confirm) |
update_campaign_status |
write | Enable / pause / remove a campaign (dry run unless confirm) |
create_ad_group |
write | Create an ad group (dry run unless confirm) |
update_ad_group_status |
write | Enable / pause / remove an ad group (dry run unless confirm) |
update_ad_group_bid |
write | Change an ad group's max CPC (dry run unless confirm) |
add_keyword |
write | Add a keyword (dry run unless confirm) |
update_keyword_status |
write | Enable / pause / remove a keyword (dry run unless confirm) |
Amounts and bids are in micros (1,000,000 micros = 1 unit of the account currency; e.g. a $5.00 daily budget is 5000000).
Development
pip install -e ".[dev]"
ruff check .
pytest -q
See CONTRIBUTING.md for the full workflow.
Security
Credentials here can change spend — treat them like production secrets. Never commit them (the .gitignore blocks key files, .env, and google-ads.yaml), grant least access, prefer a test account while evaluating, and keep the dry-run guard: review the validate_only result before re-running with confirm=True. See SECURITY.md.
Notes
- The
mcpdependency is pinned to>=1.2,<2. The 2.0 SDK reorganized its API and removedmcp.server.fastmcp, which this server uses. - This is an independent open-source project and is not affiliated with or endorsed by Google. It is separate from Google's official read-only google-ads-mcp.
License
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。