Google MCP Server
Enables appending to Google Docs and creating Gmail drafts with OAuth 2.0 authentication and an approval gate that requires manual confirmation before each action.
README
Google MCP Server
A FastAPI server that integrates with Google Docs and Gmail. Every action requires an explicit confirmation in the request before it runs, so nothing happens by accident — and this works both locally and when deployed.
Features
- Append to Google Doc — append text to any document you have access to
- Create Gmail draft — create a draft email without sending it
- OAuth 2.0 — credentials are saved locally after the first login
- Approval gate — every action requires
"confirm": truein the request body. Without it, the server returns HTTP 428 with a preview and does nothing. - API-key auth — when
API_SECRET_KEYis set, requests must include anX-API-Keyheader.
Project Structure
google-mcp-server/
├── server.py → FastAPI app with tool endpoints
├── auth.py → Google OAuth authentication
├── docs_tool.py → Google Docs tool (append content)
├── gmail_tool.py → Gmail tool (create draft)
├── requirements.txt → All dependencies
├── README.md → Setup and usage instructions
├── credentials.json → (NOT committed — downloaded from Google Cloud)
└── token.json → (NOT committed — auto-generated after OAuth)
Prerequisites
- Python 3.10+
- A Google Cloud project with the Google Docs API and Gmail API enabled
Google Cloud Setup
- Go to Google Cloud Console.
- Create a project (or select an existing one).
- Enable these APIs:
- Go to APIs & Services → Credentials.
- Click Create Credentials → OAuth client ID.
- Choose Desktop app as the application type.
- Download the JSON file and save it as
credentials.jsonin this directory.
Installation
cd google-mcp-server
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
First-Time Authentication
On the first API call (or when you run the server and trigger an endpoint), a browser window opens for Google sign-in. After you approve, a token.json file is created automatically. Subsequent runs reuse that token and skip the browser flow.
To re-authenticate, delete token.json and restart the server.
Running the Server
python server.py
Or with uvicorn directly:
uvicorn server:app --host 0.0.0.0 --port 8000
The server listens on http://localhost:8000. Interactive API docs are at http://localhost:8000/docs.
Approval model
Approval has two interchangeable modes, selected by the REQUIRE_TERMINAL_APPROVAL environment variable:
REQUIRE_TERMINAL_APPROVAL=true(local, foreground terminal) → the server prints the action and payload and promptsApprove? (y/n)in the terminal. Answering anything butyreturns HTTP 403.REQUIRE_TERMINAL_APPROVAL=false(default; required when deployed) → the caller must include"confirm": truein the request body. Without it, the server returns HTTP 428 (Precondition Required) with a preview and executes nothing.
The headless confirm mode exists because a deployed server (e.g. on Railway) has no terminal to prompt in. See DEPLOYMENT_PLAN.md for details.
API Endpoints
If
API_SECRET_KEYis configured, add-H "X-API-Key: YOUR_API_KEY"to every request below.
POST /append_to_doc
Append text to a Google Doc.
Request body:
{
"doc_id": "YOUR_GOOGLE_DOC_ID",
"content": "Text to append at the end of the document.\n",
"confirm": true
}
The doc_id is the string between /d/ and /edit in the document URL:
https://docs.google.com/document/d/ABC123xyz/edit
^^^^^^^^^^
doc_id
Example:
curl -X POST http://localhost:8000/append_to_doc \
-H "Content-Type: application/json" \
-d '{"doc_id": "ABC123xyz", "content": "Hello from the MCP server!\n", "confirm": true}'
If you omit "confirm": true, the server responds with HTTP 428 and a preview instead of running the action:
{
"detail": {
"message": "Confirmation required for 'append_to_doc'. Resend the same request with \"confirm\": true to proceed.",
"action": "append_to_doc",
"payload": {"doc_id": "ABC123xyz", "content": "Hello from the MCP server!\n"}
}
}
POST /create_email_draft
Create a Gmail draft.
Request body:
{
"to": "recipient@example.com",
"subject": "Draft subject line",
"body": "Plain-text body of the email.",
"confirm": true
}
Example:
curl -X POST http://localhost:8000/create_email_draft \
-H "Content-Type: application/json" \
-d '{"to": "recipient@example.com", "subject": "Hello", "body": "This is a draft.", "confirm": true}'
The same confirm approval applies before the draft is created.
OAuth Scopes
| Scope | Purpose |
|---|---|
https://www.googleapis.com/auth/documents |
Read and edit Google Docs |
https://www.googleapis.com/auth/gmail.compose |
Create and manage Gmail drafts |
Security Notes
- Never commit
credentials.jsonortoken.json— both are listed in.gitignore. - Every action requires
"confirm": true; requests without it return HTTP 428 and execute nothing. - Set
API_SECRET_KEYin any shared/deployed environment so requests must carry a validX-API-Keyheader.
Troubleshooting
| Issue | Fix |
|---|---|
Missing credentials.json |
Download OAuth credentials from Google Cloud and place the file in this directory. |
403 Access Not Configured |
Enable the Docs and Gmail APIs in your Google Cloud project. |
| Token expired / invalid | Delete token.json and re-authenticate. |
| Doc not found | Confirm the doc ID and that the signed-in account has edit access. |
428 Precondition Required |
Add "confirm": true to the request body to approve the action. |
401 Invalid or missing API key |
Include the X-API-Key header matching the server's API_SECRET_KEY. |
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。