Zappy MCP
Enables sending, reading, and deleting WhatsApp messages through Claude Desktop and other MCP clients with granular per-chat permissions. Built on whatsapp-web.js using a headless browser to automate WhatsApp Web.
README
zappy-mcp
WhatsApp MCP server for Claude Desktop, OpenCode, and other MCP clients. Send, read, and delete messages with granular per-chat permissions.
Built on whatsapp-web.js and the Model Context Protocol.
Table of Contents
Quick Start
git clone https://github.com/yourname/zappy-mcp.git
cd zappy-mcp
npm install
node src/index.js # Opens browser with QR code
Scan the QR with WhatsApp (Settings > Linked Devices > Link a Device).
How It Works
This server uses whatsapp-web.js, which runs a headless Chromium browser via Puppeteer to connect to WhatsApp Web. There's no official WhatsApp API for personal accounts, so this library automates the web interface.
Authentication Flow
- First run: The server launches headless Chromium and opens WhatsApp Web
- QR code: A browser window opens showing a QR code (auto-opens on an available port)
- Scan: Open WhatsApp on your phone -> Settings -> Linked Devices -> Link a Device -> Scan the QR
- Session saved: After successful scan, session credentials are stored locally
- Future runs: The server reconnects automatically using saved credentials - no QR needed
Session Storage
~/.config/zappy-mcp/
auth/ # WhatsApp session data (shared across all configs)
The auth is stored globally, so you only authenticate once per machine. Different projects can use different permission configs while sharing the same WhatsApp session.
Re-authenticating
If you need to switch accounts or fix auth issues:
rm -rf ~/.config/zappy-mcp/auth
node src/index.js # Opens QR code again
Security Notes
- Session data in
~/.config/zappy-mcp/auth/grants full access to your WhatsApp - protect it like a password - The QR code is only shown locally in your browser, never transmitted
- Each config file controls which chats the AI can access - use minimal permissions
MCP Setup
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"zappy-mcp": {
"command": "node",
"args": ["/path/to/zappy-mcp/src/index.js", "--config", "/path/to/config.json"]
}
}
}
OpenCode
Project-level (.mcp.json in project root):
{
"mcpServers": {
"zappy-mcp": {
"type": "stdio",
"command": "node",
"args": ["/path/to/zappy-mcp/src/index.js", "--config", ".zappy-mcp.json"]
}
}
}
User-level (~/.config/opencode/mcp.json):
{
"mcpServers": {
"zappy-mcp": {
"type": "stdio",
"command": "node",
"args": ["/path/to/zappy-mcp/src/index.js", "--config", "~/.config/zappy-mcp/config.json"]
}
}
}
Permissions
By default, no chats are allowed. The AI cannot send, read, or delete messages until you explicitly grant access. This is a safety feature - you control exactly which conversations the AI can interact with.
Finding Chat IDs
First, connect WhatsApp and use the list_chats tool to discover your chat IDs:
list_chats() # All chats
list_chats(groupsOnly: true) # Only groups
Chat IDs look like:
- Groups:
120363295812730408@g.us - Contacts:
14155551234@c.us(country code + phone number)
Config File
Create a config file (e.g., .zappy-mcp.json) with the chats you want to allow:
{
"allowed": [
{
"id": "120363295812730408@g.us",
"name": "Work Group",
"canSend": true,
"canRead": true,
"canDelete": false
},
{
"id": "14155551234@c.us",
"name": "Alice",
"canSend": true,
"canRead": true,
"canDelete": true
}
]
}
Permission Options
| Field | Description | Default |
|---|---|---|
id |
Chat ID from list_chats (required) |
- |
name |
Human-readable label for your reference | - |
canSend |
Allow AI to send messages to this chat | true |
canRead |
Allow AI to read messages from this chat | true |
canDelete |
Allow AI to delete its own messages (safety: off by default) | false |
Example Use Cases
| Scenario | canSend | canRead | canDelete |
|---|---|---|---|
| Full access | true |
true |
true |
| Announcements only (no reading) | true |
false |
false |
| Monitoring only (no sending) | false |
true |
false |
| Send with ability to unsend mistakes | true |
true |
true |
Global Options
At the root level of your config:
{
"suppressWarnings": true,
"allowed": [...]
}
| Option | Description |
|---|---|
suppressWarnings |
Hide "no recipients configured" warnings at startup |
Tools
Status & Discovery
| Tool | Description |
|---|---|
get_status |
Check WhatsApp connection status, shows config path and auth location |
list_allowed |
Show all permitted chats with their current permissions |
list_chats |
List all WhatsApp chats with IDs - use this to find chat IDs for setup |
Messaging
send_message
Send a message to a chat. Requires canSend permission.
| Parameter | Description |
|---|---|
to |
Chat ID (from list_chats) or phone number with country code |
message |
Text content to send |
get_messages
Fetch recent messages from a chat. Requires canRead permission.
| Parameter | Description |
|---|---|
chatId |
Chat ID or phone number |
limit |
Number of messages to fetch (default: 20) |
delete_message
Delete a message you sent. Requires canDelete permission. Only works on messages where fromMe: true.
| Parameter | Description |
|---|---|
chatId |
Chat ID where the message exists |
messageId |
Message ID (get this from get_messages response) |
forEveryone |
If true, deletes for all participants. If false, only hides it for you. (default: true) |
Troubleshooting
Messages not delivering
The WhatsApp client needs about 5 seconds after connecting to sync with the server. This is handled automatically, but if messages stay stuck in "pending" state, try restarting the MCP server.
Session/auth issues
If you're having trouble connecting, try clearing the auth data and scanning the QR code again:
rm -rf ~/.config/zappy-mcp/auth
node src/index.js
Linux: Puppeteer/Chromium dependencies
whatsapp-web.js uses Puppeteer to run headless Chromium. On Linux, you may need to install system dependencies:
sudo apt-get install -y libgbm-dev libasound2 libatk1.0-0 libcups2 libxss1 libnss3 libgtk-3-0
Account restrictions
WhatsApp does not officially support automation or bots on personal accounts. Excessive or abusive use may result in temporary or permanent account restrictions. Use responsibly and respect rate limits.
License
MIT
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。