WhatsApp MCP lets AI agents read personal chats on Cloudflare Workers

The deployment requires a Cloudflare Workers Paid plan, which the author says costs a few dollars a month, Node 20 or later, and a WhatsApp‑linked phone. The code uses the open‑source Baileys library to mimic the WhatsApp Web protocol, a method that WhatsApp’s terms prohibit. Data never leaves the user’s Cloudflare account, and media files are stored only as labels (e.g., [image], [voice note 12s]) without transcription. The author notes that the first import may be fragile on the free plan and that the server must be re‑linked if the phone removes the device.

Because the server is a personal deployment, each instance handles a single WhatsApp account and its own login credentials. The design deliberately avoids contacting external contacts services; instead, the AI agent can look up unknown numbers in the user’s own contacts and save names back into the database.