Four ways to get a model and your WhatsApp talking to each other. They differ enormously in how much context the model actually gets.
Method 1: copy and paste
Zero setup, works today, and you are the integration. Fine for a one-off. Falls apart the moment the useful context is spread across three threads and two months.
Method 2: a bot on a second number
You message a bot, the bot messages a model. It cannot see your existing chats, so it is a chat client for a model, not an assistant for your inbox.
Method 3: the Business API
Template-gated, priced per message, designed for customer messaging. Great for support automation. Structurally unable to read your personal conversations.
Method 4: MCP
Expose your WhatsApp as a set of tools over the Model Context Protocol, then point any MCP-capable client at it. The model can search and read your chats, and — if you grant write scope — reply, react, and organise, with sends still gated by your per-chat rules.
Which to pick
- One-off question about one thread: copy and paste.
- Customer support at volume: the Business API.
- You want a model that knows your actual conversations: MCP, with read-only scope first.
Start read-only. You can reconnect with a higher scope later, and you can revoke at any time.