WhatsApp MCP Server
Enables AI assistants to manage WhatsApp: list chats, send and receive messages, download media, and transcribe voice notes via the unofficial Baileys library.
README
WhatsApp MCP Server
A Model Context Protocol (MCP) server that connects to WhatsApp via Baileys, providing AI with the ability to manage WhatsApp. This is a security-hardened fork of karlfoster/whatsapp-mcp-2.0.
⚠️ WARNING
This project uses the UNOFFICIAL WhatsApp Web API (Baileys). WhatsApp may ban accounts using unofficial clients. USE A DEDICATED BURNER NUMBER — never use your personal number. The authors take no responsibility for account bans or any other consequences of using this software.
What This Is
This is a fork of karlfoster/whatsapp-mcp-2.0 with significant security hardening and operational improvements. It allows an AI assistant (like Claude) to:
- List and search chats, contacts, and messages.
- Send text messages and files (images, videos, documents, audio).
- Download media from received messages.
- Transcribe voice notes via Whisper-compatible APIs.
- Sync phone contacts from VCF files.
Setup
-
Clone the repository:
git clone <repo-url> cd whatsapp-mcp -
Install dependencies:
npm ci -
Configure environment variables: Create a
.envfile or set them in your MCP client configuration (see below). -
Run the server:
npm run dev -
Scan the QR code: On first run, a QR code will appear in your terminal. Scan it with WhatsApp on your phone: WhatsApp > Settings > Linked Devices > Link a Device
Environment Variables
| Variable | Default | Description |
|---|---|---|
ALLOWED_SEND_DIR |
./uploads/ |
Directory for outbound file sends. |
MAX_SEND_FILE_SIZE |
67108864 (64MB) |
Max file size for send_file. |
DOWNLOADS_DIR |
./downloads/ |
Directory for downloaded media. |
CONTACTS_DIR |
./contacts/ |
Base directory for VCF imports. |
WHISPER_API_URL |
(none) | Whisper API endpoint for transcription. |
WHISPER_API_KEY |
(none) | API key for Whisper. |
WHISPER_MODEL |
whisper-large-v3-turbo |
Model name for transcription. |
ZOMBIE_TIMEOUT_MS |
120000 (2min) |
Silence before zombie connection detection. |
MAX_SEND_FAILURES |
3 |
Consecutive failures before reconnect. |
MIN_SEND_INTERVAL_MS |
3000 (3s) |
Minimum delay between sends. |
SEND_JITTER_MS |
2000 (2s) |
Random jitter added to send delay. |
MAX_RECONNECT_ATTEMPTS |
10 |
Max reconnect attempts before giving up. |
Storage Locations
auth_info/: WhatsApp authentication credentials (DO NOT commit).data/: SQLite database containing messages, chats, and contacts.store/: Baileys message store and.whatsapp.lockfile.uploads/: Files to send (only files in this directory can be sent viasend_file).downloads/: Downloaded media files.contacts/: VCF files for contact import.
Running Tests
This project uses Vitest for testing.
# Run all tests
npx vitest run
# Run tests with coverage
npx vitest run --coverage
Known Limitations
- No FTS5: SQLite full-text search is not used; substring search (
LIKE) is used instead. - No Read Receipts: The server does not send read receipts.
- No Reactions: Message reactions are not supported.
- No Stories: WhatsApp Stories/Status updates are not accessible.
Security Changes (What changed from upstream)
- Path Traversal Protection: Strict containment checks for
send_fileanddownload_mediausing resolved paths. - SSRF Validation:
WHISPER_API_URLis validated to prevent SSRF attacks (rejects localhost and private IP ranges). - Atomic Lock File: Prevents multiple instances from connecting to the same WhatsApp account simultaneously.
- Strict Filename Sanitization: All filenames are sanitized to prevent injection and traversal.
- Bounded Memory Caches: Prevents unbounded memory growth in Baileys message retry and device tracking.
- Pre-key Pruning: Automatically prunes old pre-keys to prevent disk bloat.
- Removed
reply_spam: Removed the UAE-specific spam reply tool for better general-purpose use. - Streamed Uploads:
send_fileuses streams instead of loading full files into memory.
Available Tools
list_chats: List chats sorted by last activity.get_chat: Get chat details with recent messages.list_messages: Get messages from a chat.search_messages: Substring search across messages.search_contacts: Find contacts by name or phone number.get_message_context: Get messages surrounding a specific message.get_my_profile: Get your own JID and profile info.update_contact: Update a contact's display name.sync_contacts: Import phone contacts from a VCF file.send_message: Send a text message (requires confirmation).send_file: Send a media file (requires confirmation).delete_message: Delete a message (requires confirmation).delete_chat: Delete an entire chat (requires confirmation).download_media: Download media from a message to disk.transcribe_voice_note: Transcribe a voice note to text.
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。