Provides read-only access to your own WhatsApp message history as a linked device: tools to search across all chats and read messages, chat and contact details, so an agent can answer questions like "what did X say about Y" or summarize a group chat. A modified copy optionally adds an opt-in, rate-capped text-only send_message tool.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
A small personal bot: it receives WhatsApp messages through Baileys, asks an OpenAI-compatible LLM for an answer, and sends it back.
The repo began as a fork of finereli/whatsapp-mcp, whose code has since been removed
(still in git history). Only LICENSE remains from it.
Baileys is unofficial and against WhatsApp's Terms of Service. The number can be banned. Use a spare number, for personal use and experiments, not customers.
Related MCP server: WhatsApp MCP
npm install
cp .env.example .env # fill in LLM_API_KEY and ALLOWED_NUMBERS
npm start # pairing code (or QR if BOT_NUMBER is empty): WhatsApp > Linked devicesNeeds Node 20.6+. The session is stored in auth/ (git-ignored); delete it to re-link.
.env)Variable | Meaning |
| Any OpenAI-compatible endpoint. Defaults to DeepSeek. |
| Persona and style of the bot. |
| The number the bot runs on (digits, country code). Set = link with a pairing code instead of a QR. |
| Comma-separated numbers (digits, with country code) the bot answers. Empty = nobody, |
| Hourly cap across all chats (0 = no cap). |
| Hourly cap per sender (0 = no cap). Keeps one number from using the whole budget in |
| How many past exchanges per chat are sent to the model (memory only, lost on restart). |
| Random "typing" delay before each reply. |
Only direct chats are answered: groups, status and your own messages are ignored.
Senders not on ALLOWED_NUMBERS are logged, so you can copy their number into it.
With ALLOWED_NUMBERS=* anyone can spend your LLM budget; the per-sender and hourly caps bound that.
The bot has no tools, so a malicious message can at most change what it says, not what it does.
Message text is sent to the LLM provider.
npm test covers the allowlist, rate limiter and history (not a live WhatsApp connection).
License: MIT.
Let Claude or ChatGPT search, read and send your WhatsApp messages over MCP. OAuth sign-in.
Free WhatsApp MCP server for ChatGPT and Claude: read and search chats, every send you approve.
Read-only WhatsApp MCP for your AI agent: it reads the chats you choose. Try the public demo.
Ask questions across your WhatsApp inbox from Claude, ChatGPT, Cursor or any MCP client.
