Gmail MCP Server
Provides MCP clients with safe local access to read, search, and send Gmail messages using token-efficient outputs and explicit permissions.
README
Gmail MCP Server
A Model Context Protocol (MCP) server that gives MCP Clients safe, local access to your Gmail account for reading, searching, and sending mail — with explicit permissions and compact, token-efficient outputs.
Privacy note: The MCP server runs locally on your machine. Gmail data is fetched directly from Google’s APIs to your machine and is only shared with Claude when you explicitly allow a tool call.
Table of Contents
- What You Get
- Prerequisites
- 1) Create a Google Cloud Project
- 2) Enable the Gmail API
- 3) Configure the OAuth Consent Screen
- 4) Create OAuth Client Credentials (Desktop App)
- 5) Get the Code
- 6) Install Python Dependencies
- 7) First Run (Authorize and Create
token.json) - 8) Wire Into Claude Desktop (MCP Config)
- 9) Use It in Claude
- Tool Reference
- Security Tips
- Troubleshooting
- Uninstall / Remove
- License (MIT)
What You Get
- Search & list messages (up to 500 per call) using Gmail’s query syntax (
from:,subject:,has:attachment, date ranges, etc.). - Compact summaries (short-key JSON) to minimize LLM tokens and latency.
- Full message fetch with size limits, optional HTML stripping, and safe defaults.
- Thread summaries (first/last message, unread flag, counts).
- Send email (plain-text) with your explicit approval.
- Caching & batching under the hood to reduce API calls and speed up responses.
Prerequisites
- A Google account with Gmail enabled.
- Python 3.9+ installed.
- An MCP host of your choice. This guide is configured for Claude Desktop (macOS or Windows).
1) Create a Google Cloud Project
- Open Google Cloud Console and sign in:
https://console.cloud.google.com/ - Create a project:
https://console.cloud.google.com/projectcreateGive it any name (e.g., Gmail MCP). Note the selected project in the top bar.
2) Enable the Gmail API
- While your project is selected, open the Gmail API page:
https://console.cloud.google.com/apis/library/gmail.googleapis.com - Click Enable.
3) Configure the OAuth Consent Screen
- Go to the OAuth consent screen:
https://console.cloud.google.com/apis/credentials/consent - User type: choose External.
- Set Publishing status to Testing.
- Under Test users, click Add users and add the Google account(s) that will use this app (typically just your own).
- Fill in App name, User support email, and Developer contact information. Save.
Using Testing mode with Test users avoids a lengthy app verification process for personal use. The Gmail scopes requested are “sensitive,” which is expected.
4) Create OAuth Client Credentials (Desktop App)
- Go to Credentials:
https://console.cloud.google.com/apis/credentials - Click Create credentials → OAuth client ID.
- Application type: choose Desktop app (recommended for local tools).
- After it’s created, click Download JSON and save it next to
gmail_mcp.pyasclient_secret.json.
Important: Use Desktop app. “Web application” often causes redirect errors for local tools.
5) Get the Code
# clone this repo (or download ZIP and extract)
git clone https://github.com/yourname/gmail-mcp.git
cd gmail-mcp
Place your downloaded client_secret.json in this folder (the same folder as gmail_mcp.py).
The app will create token.json here after your first login.
Do not commit secrets. Add both files to
.gitignore:client_secret.json token.json
6) Install Python Dependencies
Install Requirements
macOS / Linux
python3 -V
python3 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
Windows (PowerShell)
py -3 -V
py -3 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt
7) First Run (Authorize and Create token.json)
Run the server once to complete Google sign-in and write token.json:
macOS / Linux
source .venv/bin/activate
python3 gmail_mcp.py
Windows (PowerShell)
.\.venv\Scripts\Activate.ps1
python .\gmail_mcp.py
- Your browser opens a Google login/consent page.
- Approve the requested scopes to allow the server to access your Gmail on your behalf.
- After success,
token.jsonis created next to the script. Keep this file safe and private.
If you see a redirect_uri_mismatch error, you likely created a Web application OAuth client instead of a Desktop app. Create a new Desktop app client and download it as
client_secret.json.
You can close the terminal after you see a “Starting Gmail MCP (compact mode)… ” log message.
8) Wire Into Claude Desktop (MCP Config)
Claude Desktop reads a JSON config file named claude_desktop_config.json.
Where is the config file?
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
In Claude Desktop you can also open this from Settings → Developer → Edit Config.
Minimal configuration examples
macOS
{
"mcpServers": {
"gmail": {
"command": "python3",
"args": ["/ABSOLUTE/PATH/TO/gmail_mcp.py"],
"env": {
"PYTHONIOENCODING": "utf-8"
}
}
}
}
Windows
{
"mcpServers": {
"gmail": {
"command": "python",
"args": ["C:\\\\ABSOLUTE\\\\PATH\\\\TO\\\\gmail_mcp.py"],
"env": {
"PYTHONIOENCODING": "utf-8"
}
}
}
}
After saving the config, fully quit and restart Claude Desktop. You should see the MCP hammer icon indicating the server is available.
9) Use It in Claude
Try prompts like:
- “Search my inbox for
subject:(invoice) newer_than:30dand summarize the top 20.” - “Open the full body of message ID
ABC123withstrip_html: true.” - “List threads from
from:recruiter@company.comand show unread counts.” - “Draft an email to
alice@example.comabout Tuesday’s meeting (do not send yet).”
Claude will ask for your approval before each tool call.
Tool Reference
All outputs are compact JSON (short keys, no extra whitespace) to reduce LLM tokens.
list_messages
List message IDs with an optional Gmail query.
{ "n": "<nextPageToken or ''>", "m": ["18c9...", "18cA..."] }
search_emails
Search and return minimized summaries.
{
"n": "<nextPageToken>",
"m": [
{
"i": "msgId",
"t": "threadId",
"f": "Name <gmail.com>",
"s": "Subject",
"d": "2025-08-23T10:01:02+05:30",
"sn": "Snippet",
"u": 0,
"sz": 12345
}
]
}
get_message_summary
One minimized summary by message ID (same shape as above without the list).
get_message_full
Full (still compact). Safe defaults: include_html=false, strip_html=false, max_chars=4096.
{
"i": "msgId",
"t": "threadId",
"h": { "f": "From <domain>", "to": "To...", "s": "Subject", "d": "ISO-8601" },
"bt": "text body (HTML converted when strip_html=true & no text/plain present)",
"bh": "html body or '' when include_html=false",
"sn": "Gmail snippet",
"sz": 12345
}
list_threads
List thread IDs.
{ "n": "<nextPageToken>", "m": ["threadId1", "threadId2"] }
get_thread_summary
Summary of a thread.
{
"i": "threadId",
"c": 5,
"u": 1,
"s": "Subject",
"f": "FirstFrom",
"lf": "LastFrom",
"fd": "FirstDate",
"ld": "LastDate",
"lsn": "LastSnippet"
}
Security Tips
-
Keep
client_secret.jsonandtoken.jsonprivate. Do not commit them to Git. -
To revoke access:
- Delete
token.jsonand run the server again to re-consent later, or - Visit Google account permissions and remove the app:
https://myaccount.google.com/permissions
- Delete
-
Scopes requested:
https://www.googleapis.com/auth/gmail.readonlyhttps://www.googleapis.com/auth/gmail.sendhttps://www.googleapis.com/auth/gmail.modify
Troubleshooting
Redirect URI mismatch
You probably created a Web application OAuth client. Make a Desktop app OAuth client and download it as client_secret.json.
“This app is blocked” or missing scopes Keep the OAuth consent screen in Testing and add your account under Test users.
403 / insufficient permissions
Delete token.json and run the server again to re-consent with the latest scopes.
Claude can’t see the server
- Restart Claude Desktop after editing the config.
- Double-check the absolute path and JSON syntax in
claude_desktop_config.json. - On Windows, ensure
PYTHONIOENCODING=utf-8is set as in the example.
Firewall/browser issues during auth
Allow the local browser window to open and http://localhost:<random-port> callbacks. Temporarily allow pop-ups if blocked.
Uninstall / Remove
- Remove the
"gmail"block fromclaude_desktop_config.jsonand restart Claude Desktop. - Delete
token.jsonand optionallyclient_secret.json. - Revoke the app in your Google Account permissions if desired:
https://myaccount.google.com/permissions
License (MIT)
Copyright (c) 2025
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and noninfringement. In no event shall the authors or copyright holders be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the Software or the use or other dealings in the Software.
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。