Every open-source WhatsApp MCP server that uses your personal number talks to WhatsApp through a library that implements the WhatsApp Web multi-device protocol. Two dominate: whatsmeow in Go and Baileys in TypeScript. A third, newer path uses Rust (whatsapp-rust, through baileyrs). None of them is an official API.
Who uses what
| Library | Language | MCP servers built on it |
|---|---|---|
| whatsmeow | Go | lharries/whatsapp-mcp, verygoodplugins/whatsapp-mcp, FelixIsaac/whatsapp-mcp-extended, BrOrlandi/whatsapp-mcp |
| Baileys | TypeScript | WACrew, jlucaso1/whatsapp-mcp-ts, finereli/whatsapp-mcp |
| whatsapp-rust (via baileyrs) | Rust | felipeadeildo/whatsapp-mcp |
The Go bridge pattern
lharries/whatsapp-mcp popularized a split design: a Go process keeps the WhatsApp session and writes to SQLite, and a Python process speaks MCP and reads that database. It is clean, and whatsmeow is mature (it also powers the mautrix-whatsapp Matrix bridge). The cost is two runtimes. You install Go and Python, run uv, and on Windows enable CGO because go-sqlite3 needs a C compiler. Both processes must be running for the integration to work.
The single-runtime Baileys pattern
Baileys runs in Node, so the WhatsApp session and the MCP server can live in one language. WACrew uses that to share code between its collector, web panel and MCP server; jlucaso1/whatsapp-mcp-ts does it in a single process. Setup is npm install or npx. The trade-off is that Baileys moves fast and you need to keep it updated when WhatsApp changes the protocol, the same as with whatsmeow.
What actually matters for you
- Runtime on your machine. If you live in Node, Baileys avoids Go and Python entirely. If you already run Go services, whatsmeow fits.
- Deployment. Both work in Docker. Baileys can even run inside Cloudflare Workers with shims (finereli does this); the Rust path runs in Durable Objects.
- Maintenance of the project, not the library. A well maintained server on either library beats an abandoned one. Check the last commit before you pick.
- Features above the library. Transcription, review before sending, an inbox and CRM are project choices, not library features.
Risk is the same
Both libraries link as an extra device on your account. WhatsApp's terms do not cover that, and accounts can be restricted. Behavior drives most of the risk: answering people who already talk to you, at a human pace, is the lowest-risk use. Bulk messages and cold outreach are what get numbers flagged. See WhatsApp MCP security.
FAQ
Is whatsmeow more stable than Baileys? Both are widely used and actively maintained. Stability in practice depends more on how the MCP server handles reconnects and session storage.
Can I switch library without losing my session? No. Each library stores its own session keys. You link a new device and start syncing again.
Which one does WACrew use? Baileys, isolated behind ports and adapters in src/whatsapp/, so the rest of the code does not depend on it directly. More in build a WhatsApp AI agent on Baileys.



