admob-mcp
A Model Context Protocol server for the Google AdMob API, enabling reading and writing of AdMob accounts, apps, ad units, mediation settings, and reports.
README
admob-mcp
A Model Context Protocol (MCP) server for the Google AdMob API. It gives MCP clients such as Claude Code, Cursor, Codex, and other agents a set of typed tools for reading AdMob accounts, apps, ad units, mediation settings, and reports, plus write tools for the operations the AdMob API supports. Reads run freely; every write requires your confirmation.
This project is API first. It only exposes operations that the official AdMob API actually supports, and it clearly documents the surfaces that AdMob keeps in the UI with no API.
Status
Version 0.1.0. Local stdio server with desktop OAuth. Not yet published to npm.
What it can do
Everything here is backed by the official AdMob API (admob.googleapis.com).
Read tools (safe, run without confirmation)
| Tool | Purpose |
|---|---|
admob_list_accounts |
List the signed-in AdMob publisher account. |
admob_get_account |
Get one account (publisher id, time zone, currency). |
admob_list_apps |
List apps under an account. |
admob_list_ad_units |
List ad units under an account. |
admob_list_ad_sources |
List mediation ad sources. |
admob_list_adapters |
List the adapters for one ad source. |
admob_list_ad_unit_mappings |
List ad unit mappings for one ad unit. |
admob_list_mediation_groups |
List mediation groups (supports a filter). |
admob_run_network_report |
Generate an AdMob Network report. |
admob_run_mediation_report |
Generate an AdMob Mediation report. |
admob_run_campaign_report |
Generate an AdMob campaign report (v1beta). |
Write tools (require confirmation, support dry run)
| Tool | Purpose |
|---|---|
admob_create_app |
Create an app. |
admob_create_ad_unit |
Create an ad unit. |
admob_create_ad_unit_mapping |
Map an ad unit to a mediation adapter. |
admob_batch_create_ad_unit_mappings |
Create up to 100 ad unit mappings at once. |
admob_create_mediation_group |
Create a mediation group. |
admob_update_mediation_group |
Update a mediation group (patch with update mask). |
admob_create_ab_experiment |
Start a mediation A/B experiment. |
admob_stop_ab_experiment |
Stop a mediation A/B experiment and pick a variant. |
All write tools are AdMob API v1beta methods. Google gates these behind allowlisted access, so a write may return a 403 until your AdMob account is granted access by your account manager. The server turns that 403 into a clear message.
Not supported (no AdMob API)
These AdMob surfaces have no public API, so this server does not implement them. It does not use browser automation. Use the AdMob web UI for:
app-ads.txt diagnostics, blocking controls, privacy and messaging (UMP consent messages), policy center, test devices, crawler access, change history, AdMob Labs, payments and payouts, users and roles, linked services (Firebase, Ad Manager, Analytics), app verification, and account cancellation.
Read the admob://capabilities resource for the machine-readable version of this map.
Safety model
The safety boundary is the same idea as your client asking permission before it runs a command.
- Read tools are annotated
readOnlyHint, so clients can run them without a prompt. - Write tools are annotated
destructiveHintand carry the Claude CoderequiresUserInteractionhint, which forces an approval prompt on every call and cannot be bypassed by auto accept modes. Other clients prompt on destructive tools as well. - Every write tool accepts
dryRun: true, which returns the exact request that would be sent without executing it. - Set
ADMOB_READONLY=1to disable all write tools. They are then not registered at all.
Requirements
- Node.js 18 or newer.
- A Google account with an AdMob account.
- A Google Cloud project with the AdMob API enabled and an OAuth client.
Setup
1. Enable the AdMob API and create an OAuth client
- In the Google Cloud console, enable the AdMob API for your project.
- Configure the OAuth consent screen. For personal use, set the user type to External, keep the publishing status in Testing, and add your Google account as a test user. Note that in Testing mode refresh tokens expire after 7 days; publish the app for long lived tokens.
- Create an OAuth client of type Desktop app. Note the client id and client secret.
The AdMob scopes are sensitive, not restricted. A published multi user app needs Google sensitive scope verification, but there is no annual security assessment for these scopes.
2. Install
Until this package is published to npm, clone and build it:
git clone https://github.com/jerry2247/admob-mcp.git
cd admob-mcp
npm install
npm run build
3. Authenticate
Provide your OAuth client in the environment and run the auth command. It opens a browser using a loopback redirect with PKCE and stores a refresh token.
export GOOGLE_CLIENT_ID="your-client-id"
export GOOGLE_CLIENT_SECRET="your-client-secret"
node dist/index.js auth
The refresh token is saved to ~/.config/admob-mcp/credentials.json with 0600 permissions. Run node dist/index.js logout to remove it.
4. Configure your MCP client
For Claude Code, add a project .mcp.json (see examples/mcp.json):
{
"mcpServers": {
"admob": {
"type": "stdio",
"command": "node",
"args": ["/absolute/path/to/admob-mcp/dist/index.js"],
"env": {
"GOOGLE_CLIENT_ID": "${GOOGLE_CLIENT_ID}",
"GOOGLE_CLIENT_SECRET": "${GOOGLE_CLIENT_SECRET}"
}
}
}
}
The ${VAR} values are expanded by Claude Code from your shell environment, so no secret is committed. Credentials from admob-mcp auth are read from the stored file, so you do not need to pass a refresh token. To run read only, add "ADMOB_READONLY": "1" to env. To fix the default account, add "ADMOB_PUBLISHER_ID": "pub-1234567890123456".
The same server works in other clients. Cursor and VS Code use the same mcpServers shape. Codex uses TOML, for example:
[mcp_servers.admob]
command = "node"
args = ["/absolute/path/to/admob-mcp/dist/index.js"]
[mcp_servers.admob.env]
GOOGLE_CLIENT_ID = "your-client-id"
GOOGLE_CLIENT_SECRET = "your-client-secret"
Resources and prompts
Resources:
admob://capabilities- what is API backed versus UI only or human only.admob://enums/reports- allowed dimensions and metrics for each report type.admob://account- the current account summary.
Prompts:
admob_revenue_review- guide a revenue review over a date range using the network and mediation reports.admob_setup_check- verify credentials and summarize available capabilities.
Reporting
Report tools take a date range (YYYY-MM-DD), a list of dimensions, and one or more metrics. The allowed values differ per report type and are listed in each tool description and in the admob://enums/reports resource. Network and mediation reports are generated as a stream and normalized into rows. The campaign report is v1beta only and limited to a 30 day range. Tool results include a row count and a capped set of rows to keep output small; raise limit to include more.
Environment variables
| Variable | Purpose |
|---|---|
GOOGLE_CLIENT_ID |
OAuth client id. Required. |
GOOGLE_CLIENT_SECRET |
OAuth client secret. Optional for Desktop clients but usually set. |
ADMOB_REFRESH_TOKEN |
Refresh token supplied directly, skipping the stored file. |
ADMOB_PUBLISHER_ID |
Default account, for example pub-1234567890123456. |
ADMOB_READONLY |
Set to 1 to disable all write tools. |
ADMOB_MCP_CREDENTIALS_PATH |
Override the stored credentials file path. |
ADMOB_MCP_LOG_LEVEL |
debug, info (default), warn, or error. |
Security notes
- Secrets are read from the environment. The
.mcp.jsonexample keeps them out of version control with${VAR}expansion. - The only secret stored at rest is the OAuth refresh token, written with 0600 permissions outside the repository.
- The server writes only JSON-RPC to stdout. All logs go to stderr.
- The server requests least privilege scopes:
admob.readonlyonly, plusadmob.monetizationwhen writes are enabled.
Development
npm run build # compile TypeScript to dist
npm run typecheck # type check without emitting
npm test # build then run the vitest suite
npm run format # format with prettier
npm run inspector # build then open the MCP Inspector against the server
Tests are fully mocked and do not call the AdMob API.
Known limitations and roadmap
- v1 of the AdMob API is read only. All writes are v1beta and allowlisted, so writes may return 403 until Google grants your account access.
- No hosted HTTP transport yet. v0.1 is local stdio only.
- No OS keychain storage yet. The refresh token is stored in a 0600 file.
- No browser or UI fallback for the surfaces that have no API.
License
MIT. See 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。
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。
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。
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。