Claude in your WhatsApp DMs, via a Go-Python bridge
It turns your personal WhatsApp account into a set of local MCP tools so an LLM can read, search, and send messages without touching a third-party cloud API.

What it does
A Go bridge logs into your personal WhatsApp account via the Web multi-device API and syncs messages into a local SQLite database. A Python MCP server sits on top, exposing that data to Claude (or any MCP client) through a set of read and write tools. You can search contacts, retrieve chat history with context, send text or media, and download attachments—all without a cloud intermediary.
The interesting bit
The project keeps your archive local by default and only ships data to the LLM when it explicitly calls a tool, which is a rare privacy-first stance for an LLM bridge. The README even cites Simon Willison’s “lethal trifecta” warning, openly admitting that prompt injection could still exfiltrate your private chats.
Key highlights
- Connects directly to WhatsApp using the
whatsmeowlibrary; no business API or third-party servers required. - Stores message history in a local SQLite database and gates LLM access behind explicit MCP tool calls.
- Supports two-way media: send images, video, documents, and voice messages (with optional FFmpeg conversion), plus download received media by
message_id. - Exposes a dozen tools including contact search, message context retrieval, and group messaging.
- Runs entirely offline as two local processes: a Go bridge for the WhatsApp connection and a Python server for the MCP layer.
Caveats
- Authentication requires scanning a QR code and may need re-authentication after roughly 20 days.
- Windows builds need CGO enabled for
go-sqlite3, which requires installing a C compiler toolchain. - The README explicitly warns that prompt injection could lead to private data exfiltration—the “lethal trifecta” of MCP servers.
Verdict
Worth a look if you want an AI assistant that can read and reply to your WhatsApp threads without handing your chat history to a cloud provider. Skip it if you’re not comfortable with an LLM having tool access to your personal messages, or if occasional database deletion and re-authentication sounds like a hobby you didn’t ask for.
Frequently asked
- What is lharries/whatsapp-mcp?
- It turns your personal WhatsApp account into a set of local MCP tools so an LLM can read, search, and send messages without touching a third-party cloud API.
- Is whatsapp-mcp open source?
- Yes — lharries/whatsapp-mcp is open source, released under the MIT license.
- What language is whatsapp-mcp written in?
- lharries/whatsapp-mcp is primarily written in Go.
- How popular is whatsapp-mcp?
- lharries/whatsapp-mcp has 6.3k stars on GitHub.
- Where can I find whatsapp-mcp?
- lharries/whatsapp-mcp is on GitHub at https://github.com/lharries/whatsapp-mcp.