How Claude reads your WhatsApp chats

Frederic Bodin · 27 October 2026

Claude has no WhatsApp connector, but on a Mac Claude Code can read the WhatsApp desktop app's own message store on your laptop, with no bridge, no API key and no third-party server. On Windows you export the chats you want from your phone and drop them in the folder.

Mac

Elena, my AI chief of staff, has run my business from one folder on my Mac since 22 July 2026. By 25 October she'd logged 1,629 jobs. 11,542 mails from six inboxes, 3,082 LinkedIn conversations, 988 WhatsApp chats and 94 Telegram chats run through one queue, read three times a day at 07:30, 13:00 and 18:00. The WhatsApp part needs nothing from WhatsApp except the app you already have.

When you link WhatsApp's desktop app, your phone sends it a copy of your recent message history, and WhatsApp says that copy is stored locally. So it's a file on your Mac. Claude Code, in the Code tab of the Claude desktop app, reads it the way it reads any other file on your laptop. It copies the store first and reads the copy, so WhatsApp's own file is never opened for writing.

The first time, macOS asks whether Claude may read another app's data. You click Allow… and that's the setup. No login, nothing to scan, no extra linked device on your account. Anthropic's own iMessage plugin for Claude Code works on the same principle with Apple's Messages: it reads the local database directly.

What it reads

The text of your chats, one-to-one and groups: who wrote each line, when, and in which chat. A group lists under its members, so a client you only ever speak to in a project group still turns up as a person, with a last word and a date.

It reads what the desktop app holds. WhatsApp says plainly that not every message and chat syncs from your phone to a linked device, and that the desktop app syncs more history than WhatsApp Web. Your recent history is there. A chat from four years ago may only be on your phone.

Voice notes, photos and documents stay where they are. Status updates are skipped.

What it does

• One list of everyone. Day 4 matches each person across mail, LinkedIn and WhatsApp and writes the 20 who matter, each with every channel they were seen on and the last word. Mail comes in through Claude's Gmail connector, LinkedIn through its own data export.
• The history before a call. Everything you and that person ever said, on one page, before you join.
• Search by meaning. "What did we agree with her in May?" Day 5 builds a semantic index on your laptop, so the answer comes back from a WhatsApp chat even when you've forgotten the exact words.
• A reply in your words. Claude drafts it from the thread and your voice file. You read it, paste it into WhatsApp and press send.

That last step is mine in my own system too. Every reply waits for my press… except a few narrow cases I switched on myself.

Limits

It doesn't send. The course gives Claude no way to send a WhatsApp message, so a draft waits in one file in your folder until you send it yourself.

It reads what's on the laptop, when the laptop is on. The three daily runs are scheduled tasks in the Claude desktop app, and Anthropic's docs say they run only while the app is open and the computer is awake. A run your laptop sleeps through is skipped, and the next one catches up.

WhatsApp logs a linked device out if your phone goes unused for more than 14 days. Open WhatsApp on your phone now and then.

Windows

On Windows the course uses WhatsApp's own export.

1. On your phone, open a chat that matters.
2. On an iPhone, tap the contact's or group's name, then Export chat, then Without media. On Android, open the menu, then More, then Export chat, then Without media.
3. Save the file or mail it to yourself, and drop it in your folder's inbox. The kit reads the ZIP as WhatsApp makes it.

Ten chats is plenty to start: your clients and your warmest leads. An export is a copy of that moment, so new messages mean a new export. Everything else works the same way: the one list, the history before a call, the search and the drafts. The same route covers a Mac without the desktop app.

Privacy

WhatsApp's privacy policy says your messages are typically stored on your devices, and deleted from its servers once they're delivered. This route keeps them on yours. Nothing logs in to your account, nothing new gets linked, and no third-party server holds a copy. You never type a password or hand anyone your number. The semantic index runs on your laptop too, in one file in your folder.

What leaves the laptop is what a task needs. Claude Code runs locally and sends your prompts and its answers to Anthropic to do the work, so the lines of a chat it reads for a draft go there, as with anything you ask Claude. On Pro and Max, Anthropic gives you the choice whether your data may be used to improve future models. I never see any of it.

Routes

Search for Claude and WhatsApp and you'll find bridges and MCP servers on GitHub. The ones that come up first link your personal account as a new device, one through whatsmeow and one through Baileys, third-party libraries for WhatsApp's web protocol. WhatsApp's help page is blunt about that kind of link: "Linking your account to an unofficial app or website, now or in the past, may result in a temporary or permanent account ban." My business runs on that number.

Claude's connector directory, searched for WhatsApp on 27 October 2026, shows three third-party tools: two platforms for messaging customers and a sales CRM. None comes from WhatsApp or Meta. And if you want to text Claude from your phone, Claude Code's channels cover Telegram, Discord and iMessage in a research preview. WhatsApp isn't on the list.

Questions

Day 4 of the course sets this up on your own laptop, Mac or Windows.

Start for €29