- Published on
Show HN: WhatsApp MCP running free on Cloudflare
- Authors
- Name
- DeepHorus
A new open-source server called WhatsApp MCP lets AI agents read and search a user’s WhatsApp messages running in a Cloudflare account. Created by finereli, it links a phone once like WhatsApp Web, then allows MCP clients like Claude Code to query chats. The server cannot send messages, keeping the phone authoritative.
The project fills a gap: WhatsApp offers no official API for reading personal messages. It relies on Baileys, an open-source library using WhatsApp Web’s protocol. This violates WhatsApp’s terms, though the developer says the practical risk for a read-only device is low but not zero.
Setup requires a Cloudflare Workers Paid plan, Node 20 or later, and a phone. Users clone the GitHub repository, run npm install, create a KV namespace, set a 12-character password, and deploy. The paid plan, a few dollars monthly, is needed because importing a large history exceeds Workers’ default CPU limit.
Messages are stored in a SQLite database inside a Durable Object in Cloudflare. The server doesn't stay connected to WhatsApp; it connects when an agent requests data and once daily to collect queued messages. Media is not downloaded, appearing only as labels like [image] or [voice note 12s].
Users can ask questions like “What did Dana say about the flights?” or “Summarize what I missed in the family group this week.” The agent can name unnamed chats by looking up numbers through existing contacts tools. The server never connects to a contacts service itself.
Because each deployment holds a message history and a live login, the tool is designed for one person: one WhatsApp account and one owner. First imports may be incomplete, and relinking is required for re-import. For now, the project offers a privacy-focused but unofficial way to bring personal chat data into AI workflows.
