WhatsApp MCP
Enables MCP clients to list, read, search, and send WhatsApp messages via a persistent WebSocket connection with local SQLite storage.
README
WhatsApp MCP
A Baileys-based MCP server for WhatsApp:
a persistent connection that maintains a local message store, exposing tools to list,
read, search, and send — so an MCP client (Claude, or an agent like garvis) can work
with your WhatsApp messages.
Unlike scraping WhatsApp Web in a browser, this speaks WhatsApp's WebSocket multi-device protocol directly: no browser, no DOM selectors, no headless crashes.
Status
🚧 Early/local development. Not yet published.
How it works
WhatsApp's multi-device protocol is event-driven and only delivers messages while you're connected, so this runs as a resident connection + local store:
- Daemon — holds the WebSocket connection, persists incoming messages to a local
SQLite store, and auto-reconnects. Auth is a paired session (
useMultiFileAuthState), created once via QR / pairing code. - MCP server — reads the store and sends via the socket, exposing the tools.
History reflects what has been synced since you connected (on-connect sync + live events), not your entire WhatsApp past.
⚠️ Important caveats
- Unofficial. This uses a reverse-engineered protocol library and is against WhatsApp's Terms of Service. Accounts can be banned, especially for high-volume or spammy automation. Use a number you can afford to lose; for sanctioned automation use the WhatsApp Business Cloud API instead.
- Personal use only, low volume, people you know.
- Deletion is guarded, never automatic: it requires an explicit
confirm: true(otherwise the tool only previews), defaults to delete-for-me, and only allows delete-for-everyone on your own messages. See the security model.
Tools
All tools are prefixed whatsapp_ to avoid name collisions with other MCP servers.
| Tool | What it does |
|---|---|
pairing_status |
Whether the session is paired and connected |
list_conversations |
Recent chats (name, last-message snippet, unread) |
read_conversation |
Recent messages in a chat (each with an id for deletion) |
send_message |
Send a text to a chat / number |
search_messages |
Search the local store |
delete_message |
Delete one message — for-me by default; for-everyone (own msgs) opt-in |
delete_conversation |
Delete a whole conversation for-me (never affects the other person) |
Delete semantics. "For me" clears the message/chat from your devices only. "For everyone" (messages only, your own only) revokes it for the recipient and leaves a visible "this message was deleted" marker. Both require
confirm: true; without it the tool returns a preview and does nothing. Deletes are applied to the local store optimistically; a long-running daemon may re-sync a deleted-for-me item if WhatsApp re-delivers it.
License
GNU AGPL-3.0-or-later. Note the network-use clause: running a modified version as a network service obligates you to offer its source to users.
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。