lharries/whatsapp-mcp made the idea popular. Connect your WhatsApp to Claude through the Model Context Protocol, and let the agent search and answer your chats. It is still the most starred project in the space, and many people searching for a WhatsApp MCP server land there first. There are now several alternatives, built on different stacks and with different goals.
This page compares the open-source options with public data from GitHub and each project's README, checked on November 14, 2026. Star counts move every day; the dates and licenses are the useful part.
The short version
| Project | Connects through | Stack | License | Last commit |
|---|---|---|---|---|
| lharries/whatsapp-mcp | WhatsApp Web (whatsmeow) | Go bridge + Python MCP | MIT | July 2025 |
| FelixIsaac/whatsapp-mcp-extended | WhatsApp Web (whatsmeow) | Go bridge + Python MCP | MIT | November 2026 |
| verygoodplugins/whatsapp-mcp | WhatsApp Web (whatsmeow) | Go bridge + Python MCP | MIT with fork notice (GitHub shows Other) | November 2026 |
| whatzmcp/whatzmcp | WhatsApp Web | Node 20+, npx or Docker | MIT | October 2026 |
| felipeadeildo/whatsapp-mcp | WhatsApp Web (whatsapp-rust) | Cloudflare Workers, remote MCP | GPL-3.0 | November 2026 |
| finereli/whatsapp-mcp | WhatsApp Web (Baileys), read-only | Cloudflare or hosted beta | MIT | November 2026 |
| BrOrlandi/whatsapp-mcp | WhatsApp Web (whatsmeow) | Go, VPS installer | MIT | October 2026 |
| jlucaso1/whatsapp-mcp-ts | WhatsApp Web (Baileys) | TypeScript | ISC in package.json, no license file | January 2026 |
| msaelices/whatsapp-mcp-server | GreenAPI (hosted) | Python | MIT | April 2025 |
| codespar/mcp-dev-latam | Z-API, Evolution API or Cloud API | TypeScript, on npm | MIT | November 2026 |
| WACrew | WhatsApp Web (Baileys) | TypeScript, on npm as wacrew-mcp | Apache 2.0 | November 2026 |
Two different ideas sit in this table. Most servers are bridges: they expose WhatsApp to the agent as tools and stop there. A few, like the codespar packages, are API wrappers that need a paid or self-hosted WhatsApp API behind them. WACrew is a full app with a review layer and a web panel, and the MCP server is one of its interfaces.
lharries/whatsapp-mcp
The original. A Go bridge connects to your personal account through the WhatsApp Web multi-device API, using the whatsmeow library, and stores messages in a local SQLite database. A Python MCP server reads that database and gives Claude tools to search messages and contacts and to send text and media. Audio as voice notes needs FFmpeg.
Its README carries an honest warning: like many MCP servers, it is exposed to the "lethal trifecta", meaning prompt injection could lead to private data leaking. That applies to every server on this page that reads your messages and can also send.
The last commit on the default branch is from July 2025, so issues and WhatsApp protocol changes since then are not reflected in the main repo.
FelixIsaac/whatsapp-mcp-extended
A maintained continuation of the same architecture (Go and whatsmeow, Python MCP, SQLite). Its README lists 26 tools by default and a lean mode with 15, plus webhooks and group management, and it ships with Docker. If you already run lharries/whatsapp-mcp and want updates without changing stacks, this is the closest drop-in.
verygoodplugins/whatsapp-mcp
Another maintained fork of the lharries design, requiring Go 1.26+ and Python 3.11+. It adds webhooks, call history, quoted replies, mentions and read receipts. Side by side with WACrew: verygoodplugins/whatsapp-mcp vs WACrew.
whatzmcp
A single Node package you run with npx -y github:whatzmcp/whatzmcp or as a Docker image, with 13 tools and send limits that ramp up from 50 messages on day one. See whatzmcp vs WACrew.
felipeadeildo/whatsapp-mcp
Runs in your own Cloudflare account (Workers Paid, about $5 a month) and connects to Claude or ChatGPT as a remote MCP server, with voice notes transcribed by Workers AI. GPL-3.0. See felipeadeildo/whatsapp-mcp vs WACrew.
finereli and BrOrlandi
finereli/whatsapp-mcp is read-only by design and stores messages encrypted. BrOrlandi/whatsapp-mcp ships a guided VPS installer in Portuguese, and BrOrlandi/whatsapp-mcp-local is a desktop app that serves MCP on localhost.
jlucaso1/whatsapp-mcp-ts
A TypeScript rewrite on Baileys instead of whatsmeow, with messages in SQLite through node:sqlite. It needs Node 23.10 or newer. Useful if you prefer one language for the whole stack. Its package.json declares ISC, but there is no license file in the repository, so GitHub does not show one; worth checking if you plan to build on it.
API-backed servers: GreenAPI, Z-API, Evolution API, Cloud API
These servers do not connect to WhatsApp themselves. They call an API you already have:
- msaelices/whatsapp-mcp-server wraps GreenAPI, a hosted service with its own account.
- codespar/mcp-dev-latam publishes separate npm packages for Z-API (
@codespar/mcp-z-api), Evolution API (@codespar/mcp-evolution-api) and the official WhatsApp Cloud API (@codespar/mcp-whatsapp-cloud), as part of a larger set of Latin American commerce servers.
This route makes sense when the WhatsApp API is already part of your stack, for example a company sending notifications through the Cloud API, and you want an agent on top of it.
WACrew
WACrew is an open-source WhatsApp AI agent with a web panel. Its MCP server (wacrew-mcp on npm, io.github.sumioshi/wacrew in the official MCP Registry) gives Claude Code, Codex, Cursor and other clients 36 tools over the same data the panel uses.
What it adds over a bridge:
- Groups first. Every group and direct chat lands in a queue sorted by waiting time, so the agent can ask "who is waiting" instead of scrolling.
- Media. Voice notes are transcribed, and images, videos and PDFs can be read.
- A review before sending. Code rules reject replies that read like a bot, repeated bubbles and over-apologizing. The generic part is the open-source botsmell linter.
- Human-paced delivery. Read receipt, typing and pauses, and the delivery stops if the contact writes again or you answer from your phone.
- Per-chat control. Confirm mode, the default, shows you every message before it goes out.
More detail on the WhatsApp MCP server page, the full list of 36 tools and the direct comparison with lharries/whatsapp-mcp.
Which one to pick
- You want the classic setup and do not mind an unmaintained main repo: lharries/whatsapp-mcp.
- Same setup, maintained: FelixIsaac/whatsapp-mcp-extended or verygoodplugins/whatsapp-mcp.
- One
npxcommand, local: whatzmcp. - Always on without a VPS: felipeadeildo/whatsapp-mcp.
- Search only, never send: finereli/whatsapp-mcp.
- You already pay for Z-API or run Evolution API or the Cloud API: the codespar packages.
- You answer clients on WhatsApp, often in groups, and want the agent to write like you and wait like a person: WACrew.
For picks by use case, see best WhatsApp MCP servers, and for the security side, WhatsApp MCP security.
The risk that applies to all of them
Every server here that connects through WhatsApp Web, whether whatsmeow or Baileys, links to your account as an extra device, outside the official Business API. WhatsApp's terms do not cover that use, and a number can be restricted. Behavior drives most of the risk: bulk sending and cold messages to strangers stand out, while answering people who already talk to you at a human pace is the lowest-risk use. For volume or official support, use the Cloud API.
FAQ
Is lharries/whatsapp-mcp still maintained? Its last commit on the default branch is from July 2025. Maintained forks and rewrites exist, listed above.
Does any WhatsApp MCP server use the official API? Yes, through wrappers such as @codespar/mcp-whatsapp-cloud. You need a Cloud API setup with Meta behind it, and group chats work differently than in the app.
Can I use these with Claude Code, Cursor or Codex? They are all MCP servers, so any client that supports their transport works; most run over stdio, and msaelices/whatsapp-mcp-server also serves HTTP. WACrew also has an AI agent guide for setups without a coding agent.
More in the comparisons index.



