Reverse engineer anything with agents, from app behavior down to native binaries.
Whatsapp MCP Kit
Link WhatsApp to Claude (or any MCP client) in ~20 minutes: a live self-refreshing QR page that makes pairing actually work, production patches, and 12 field gotchas. On top of…
Install
claude mcp add whatsapp --scope user -- node "$(npm root -g)/@sjawhar/whatsapp-mcp/dist/index.js"Install command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →whatsapp-mcp-kit
Connect Claude (Claude Code / Claude Desktop / Codex — any MCP client) to a WhatsApp account so it can read your chats and send messages. The server already exists and is good; what nobody ships is the linking procedure. Ours took an evening of broken iterations and a re-scan loop; with this kit it should take you ~20 minutes.
Built and battle-tested at Palo Alto AI Research Lab — our Claude fleet has been reading WhatsApp through exactly this setup since June 2026.
⚠️ Read
docs/SECURITY.mdfirst. Baileys is an unofficial WhatsApp client; upstream recommends a dedicated number, not your personal one. We linked a main number knowingly and accept the ban risk. That should be a decision, not an accident.
What you get
- A working setup path on top of the upstream server
@sjawhar/whatsapp-mcp(Node/Baileys, security-hardened fork ofkarlfoster/whatsapp-mcp-2.0; 17 tools — list/search chats, messages, contacts, send text & files, download media, transcribe voice notes). PROMPT.md— a copy-paste prompt you hand to Claude Code or Codex; it performs the installation and walks you through the QR scan, gotchas included.pair/wa_qr_live.py— the piece that made linking actually work: it keeps the stdio server alive, scrapes the raw QR string out of its log, renders a PNG and serves a live localhost page that refreshes the code by itself. Plus a 27-check regression test that needs no phone, no network and no server installed.- Patches we run in production, pinned to upstream npm v2.4.0 (
git apply --checkproven on a freshnpm packunpack):patches/0001-qr-raw-to-stderr.patch— print the raw QR string, always. Upstream only emits terminal ASCII art (and the raw string solely in a.catch()fallback), and that art is unscannable mosaic in most Windows console fonts. Without this patch there is nothing to render a PNG from.patches/0002-group-subject-resolve.patch— real group names.db.getChatalways fills a<digits> (group)placeholder, soresolveChatNamenever fires and every group shows up as a number. Cost us ~70 unnamed groups.
docs/GOTCHAS.md— every trap we paid hours for, with dates.docs/SECURITY.md— a WhatsApp MCP is your whole WhatsApp. Read before linking.
Quickstart (manual)
# 0. get the kit (patches/ and pair/ must sit together on disk)
git clone https://github.com/tonydzi/whatsapp-mcp-kit ~/whatsapp-mcp-kit
# 1. install the server (Node >= 18)
npm i -g @sjawhar/whatsapp-mcp
# 2. apply the kit patches to the installed package
# (the folder is not a git repo -- git apply patches a plain working tree just fine)
cd "$(npm root -g)/@sjawhar/whatsapp-mcp"
git apply ~/whatsapp-mcp-kit/patches/0001-qr-raw-to-stderr.patch
git apply ~/whatsapp-mcp-kit/patches/0002-group-subject-resolve.patch
# 3. link the phone: live QR page on http://127.0.0.1:8799
pip install qrcode pillow
python ~/whatsapp-mcp-kit/pair/wa_qr_live.py
# 4. register in Claude Code
claude mcp add whatsapp --scope user -- node "$(npm root -g)/@sjawhar/whatsapp-mcp/dist/index.js"
Similar plugs
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype…
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io