Gmail MCP Server (Multi-User)
Multi-user Gmail MCP server that enables sending/receiving emails, managing accounts, and searching mail via SMTP/IMAP with per-user isolation and encrypted app passwords.
README
Gmail MCP Server (Multi-User)
A Model Context Protocol (MCP) server for Gmail that supports many different users, each connecting their own Gmail account. Built in Python with FastMCP and FastAPI,extended with per-user account management.
Each user authenticates with their own Google App Password over standard SMTP/IMAP (no OAuth consent flow needed). Passwords are encrypted at rest and never returned by any tool or endpoint.
How multi-user works
Every registered Gmail account is a row tied to a user_id that you supply (your app's own user identifier — this server doesn't do authentication for you). All contacts, conversation threads, and messages are scoped to account_id, so different users' mailboxes never mix. One deployment can serve any number of users, each with one or more Gmail addresses connected.
Features
- Multi-user accounts:
add_gmail_account/list_gmail_accounts/remove_gmail_account. - Encrypted credentials: App passwords encrypted at rest with
cryptography.Fernet. - Database persistence: Contacts, threads, and messages stored per-account in SQLite.
- Background inbox polling: Since Gmail has no webhook push over IMAP, a background task polls every active account's inbox on an interval and imports new mail automatically.
- FastMCP Tools:
add_gmail_account/list_gmail_accounts/remove_gmail_accountsend_email: Sends plain-text/HTML email with optional CC/BCC/attachments, from a specific account.fetch_inbox: Polls IMAP right now for a specific account.get_emails: Retrieves chronological message history with a contact.search_emails: Keyword search across subject/body.get_contact/list_conversations
- FastMCP Resources (Read-Only):
conversations://{account_id}contacts://{account_id}messages://{account_id}/{email}
- REST endpoints:
POST/GET /accounts,DELETE /accounts/{id},POST /accounts/{id}/fetch,GET /health.
Project Structure
gmail-mcp-multiuser/
├── app.py # Entry point (FastAPI + FastMCP mounting + poller)
├── config.py # Pydantic Settings env loader (server-wide only)
├── requirements.txt
│
├── mcp_server/
│ ├── server.py # FastMCP instance
│ ├── registry.py # Imports all tools/resources to register them
│ ├── tools/
│ │ ├── accounts.py # add/list/remove Gmail accounts (multi-user)
│ │ ├── messages.py # send_email, fetch_inbox, get_emails, search_emails
│ │ └── contacts.py # get_contact, list_conversations
│ └── resources/
│ └── gmail_resources.py # Read-only resource handlers
│
├── gmail/ # Gmail SMTP/IMAP wrappers (fixed Gmail endpoints)
│ ├── smtp_client.py # Sends mail per-account via SMTP
│ ├── imap_client.py # Fetches mail per-account via IMAP
│ ├── parser.py # Parses raw RFC822 bytes into structured data
│ └── crypto.py # Fernet encryption for stored app passwords
│
├── services/
│ ├── account_service.py
│ ├── message_service.py
│ ├── contact_service.py
│ └── conversation_service.py
│
├── database/
│ ├── database.py
│ ├── models.py # GmailAccount, Contact, Conversation, Message
│ └── crud.py
│
├── routes/
│ ├── accounts.py # REST account management + manual fetch trigger
│ └── health.py
│
└── utils/
├── logger.py
└── validators.py
Getting Started
1. Installation
cd gmail-mcp-multiuser
python -m venv .venv
# Windows:
.venv\Scripts\Activate.ps1
# macOS/Linux:
source .venv/bin/activate
pip install -r requirements.txt
2. Environment Configuration
cp .env.example .env
Generate a Fernet encryption key and put it in .env:
python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
Fill in .env:
ENCRYPTION_KEY: the generated Fernet key above (required — encrypts every user's stored app password).DATABASE_URL: defaults to local SQLite, fine for development.INBOX_POLL_INTERVAL_SECONDS/ENABLE_BACKGROUND_POLLING: control automatic inbox polling.
Note: there's no single Gmail account here — each user registers their own in step 4.
3. Run the server
python app.py
Exposed endpoints:
GET http://localhost:8000/healthPOST/GET http://localhost:8000/accountsPOST http://localhost:8000/accounts/{id}/fetchGET/POST http://localhost:8000/mcp/sse— FastMCP Server-Sent-Events endpoint for MCP clients
4. Register a user's Gmail account
Each user needs a Google App Password (not their normal password):
- Enable 2-Step Verification: https://myaccount.google.com/security
- Generate an App Password: https://myaccount.google.com/apppasswords
Then call the add_gmail_account MCP tool (or POST /accounts):
{
"user_id": "user-123",
"email": "jane@gmail.com",
"app_password": "abcdefghijklmnop",
"display_name": "Jane Doe"
}
This returns an account_id — use that for every subsequent send_email, fetch_inbox, get_emails, etc. call for this user.
5. Send and receive
send_email(account_id, to, subject, body, ...)sends immediately via SMTP.- New inbound mail is imported automatically by the background poller, or on demand via
fetch_inbox(account_id)/POST /accounts/{id}/fetch.
Notes
- App passwords are encrypted at rest but never logged or returned by any tool/endpoint.
- Users are isolated purely by
account_idscoping in the database — there is no built-in authentication layer for your application's end users; add one in front of this service if it's exposed beyond trusted internal callers. - If
fetch_inbox/send_emailreturn an authentication error, the most common cause is using a normal Gmail password instead of an App Password, or IMAP being disabled in Gmail settings (Settings → Forwarding and POP/IMAP → Enable IMAP).
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。