Sitemap

Baileys, Evolution API, whatsapp-web.js, and wppconnect all speak WhatsApp

WAHA lets you pick an engine: WEBJS, NOWEB, or GOWS. All three are reverse-engineered WhatsApp clients that pair a linked device and sit on Meta

Most comparisons frame this as personal vs verified business. The real decision axis is wall-clock throughput. Desktop automation hits a hard ceiling near 15 messages per minute on one Mac, measured from the actual Thread.sleep timings in handleSendMessage. The Cloud API has no equivalent ceiling but asks for verified-business onboarding and async webhook handling. Here is the latency budget on both sides and the math that decides for you.

Compare the two paths by what an AI agent actually sees through each. WhatsApp MCP exposes 11 named tools, six of them read-side against your existing contacts and chat history. The Business API exposes graph endpoints that have no read access to your contact directory at all. That asymmetry is what makes the choice obvious for personal-account agents.

Common questions about the WhatsApp MCP server on macOS: how it works, ban risk, supported clients, and the differences vs the WhatsApp Business API or WhatsApp Web automation.

Install the WhatsApp MCP server on macOS. Single npm command, Swift postinstall, then register the server in Claude Desktop, Claude Code, Cursor, or any MCP client.

15 minutes with the maintainer of WhatsApp MCP for macOS. Talk through your automation idea, agent setup, or integration question.

What WhatsApp MCP for macOS does and does not do with your data. Local-only execution, no telemetry by default, optional analytics on this marketing site.

Every page on whatsapp-mcp-macos.com, grouped by section.

Per WhatsApp

Every guide tells you to call AXUIElementCreateApplication and walk kAXChildrenAttribute. None of them tell you what WhatsApp Catalyst actually exposes. Here are the five VoiceOver description patterns and three spatial thresholds whatsapp-mcp-macos matches against, lifted directly from the Swift source.

Guides for the WhatsApp MCP server on macOS — installation, MCP tooling, npm package management, and adjacent automation topics.

Manychat drives WhatsApp through Meta

Every

Most guides treat the description string on an MCP tool as documentation. On a shipping WhatsApp MCP server, the description is a micro-prompt that names the next tool to call, the verification step in between, and what NOT to do. Traced through Sources/WhatsAppMCP/main.swift.

`npm-check

OpenSearch MCP server is the canonical example of a stateless MCP server. Every tool call is self-contained because the data is keyed by document id. WhatsApp MCP for macOS is the opposite: its search tool leaves the UI open and the next call consumes that state by index. Two designs, side by side.

openwa drives WhatsApp Web through Puppeteer and injects a script that pulls Meta

The famous risk of a self-hosted WhatsApp gateway is the account ban. The hidden one is custody: the gateway stores your full WhatsApp session as a portable plaintext file and runs it through an unaudited npm dependency tree. lotusbail, a poisoned Baileys clone with 56,000+ downloads, exfiltrated exactly that credential before it was caught in December 2025.

Wiring an AI agent to WhatsApp means picking TWO transports stacked: stdio JSON-RPC between the agent and the MCP server, then a WhatsApp-side transport (Meta Cloud HTTPS, whatsmeow WebSocket, browser CDP, or macOS accessibility). The second one is where the real architectural choice lives.

No official WhatsApp API works without a Meta Business account. The unofficial APIs that skip Meta link a reverse-engineered client device and get numbers banned in 2 to 8 weeks. A third path drives the genuine WhatsApp Desktop app via macOS accessibility and never connects to WhatsApp servers as a client at all.

You can automate WhatsApp without speaking its web protocol. Instead of opening a WebSocket session like Baileys or running WhatsApp Web in a headless browser, whatsapp-mcp-macos reads the macOS accessibility tree of the genuine desktop app and posts synthetic clicks. No new linked device, no reverse-engineered client.

Booking automation on WhatsApp usually means Calendly + Zapier + a Meta Business API template firing into the void. The desktop-client path is different: a local MCP server drives the WhatsApp Mac app, reads the existing booking thread, drafts a reply that references the customer

Meta

Every other WhatsApp MCP that skips Meta

A WhatsApp MCP server exposes WhatsApp as JSON-RPC tools an AI can call. The four real implementations differ in HOW they reach WhatsApp: Business Cloud API, whatsmeow web-multidevice, headless WhatsApp Web, or macOS accessibility. Side-by-side, with the failure modes each one hides.

Most guides on automating orders from WhatsApp voice notes route you through Twilio, the Business Cloud API, and a Whisper sidecar. There is a shorter path on macOS: WhatsApp