WACrew

Guide

Open-source WhatsApp AI agents on GitHub, and how to pick one

Open-source WhatsApp AI agent projects on GitHub in 2026, from MCP servers to gateways and libraries, with license, last activity and a checklist to pick one you can run in production.

Updated 3 min read
On this page
  1. The three layers
  2. Projects, as of 2026-11-17
  3. A checklist before you clone
  4. Where WACrew fits
  5. FAQ

Searching for a WhatsApp AI agent on GitHub returns three different kinds of project under the same words: MCP servers that let an AI agent use WhatsApp, HTTP gateways you build an agent on, and the low-level libraries underneath both. Mixing them up is how people end up with a library when they wanted a working agent, or a gateway when they wanted Claude to answer their clients.

The three layers

MCP servers connect WhatsApp to an agent you already use (Claude Code, Codex, Cursor, Claude Desktop). You ask in plain language and the agent calls the tools. Pick one if the AI part already exists and WhatsApp is what is missing.

Gateways hold the WhatsApp session and expose REST or webhooks. You write the agent loop yourself, in any language. Pick one if you are building a product.

Libraries are the protocol layer. Everything above is built on one of them.

Projects, as of 2026-11-17

projectlayerlanguagelicenselast push
sumioshi/wacrewMCP server + collector + panel + REST APITypeScriptApache 2.02026-11
lharries/whatsapp-mcpMCP serverGo + PythonMIT2025-07
verygoodplugins/whatsapp-mcpMCP serverGosee repo2026-11
felipeadeildo/whatsapp-mcpMCP serverTypeScriptGPL-3.02026-11
evolution-foundation/evolution-apigatewayTypeScriptsee repo2026-11
devlikeapro/wahagatewayTypeScriptApache 2.02026-11
WhiskeySockets/BaileyslibraryJavaScriptMIT2026-11
tulir/whatsmeowlibraryGoMPL-2.02026-11

lharries/whatsapp-mcp is the best known WhatsApp MCP server (about 6.9k stars), but its last push was in July 2025. We compared it, and the others, in detail: lharries alternatives, verygoodplugins vs WACrew, felipeadeildo vs WACrew and WAHA vs Evolution API.

A checklist before you clone

  1. Last commit and open issues. WhatsApp changes its protocol; a project that stopped in 2025 may already be broken.
  2. License. GPL-3.0 changes what you can ship in a closed product; MIT and Apache 2.0 do not.
  3. Where the session lives. Your laptop (stops when it sleeps) or a server (24/7, but you host it).
  4. History. Does it store messages, or only see what arrives while running? An agent that cannot search last week is half an agent.
  5. Voice notes and images. Clients send audio. Without transcription the agent is blind to half the conversation.
  6. Sending safety. Can the agent send instantly with no review? Look for a confirm mode, a cancel window and pacing.
  7. Setup cost. Go toolchain, Python, Docker, or a single npx.

Where WACrew fits

WACrew covers the MCP and gateway layers in one project: a collector on Baileys that stores history, a web panel with the inbox sorted by wait time, an MCP server with 37 tools (npx -y wacrew-mcp, also in the official MCP Registry) and a REST API. Replies pass a deterministic review against robotic text and leave at a human pace, each chat is confirm or auto, and gerar_voz measures how you write so the agent can copy it. It is young (9 stars at the time of writing), so weigh that the same way you weigh an old project's last commit.

FAQ

Is there an official WhatsApp AI agent on GitHub? Meta does not publish one. Projects on GitHub are community built and, apart from those using the Cloud API, connect as a linked device.

Which one works with Claude Code? Any MCP server. WACrew, lharries/whatsapp-mcp, verygoodplugins and felipeadeildo all speak MCP; they differ in setup, history and sending safety.

Can I self-host all of these? Yes, all the projects in the table are self-hosted. None of them is a SaaS you sign up for.

Do these agents risk my number? Reading and answering people who write to you is low risk. Bulk and cold messaging is what gets numbers banned.