About Agent Messenger

Agent Messenger lets an AI assistant such as Claude read and search your WhatsApp conversations, and send messages when you allow it, without handing it unsupervised reach into real customer and personal conversations.

How it works

You link a WhatsApp number the same way you would link WhatsApp Web, by scanning a QR code. You then connect an assistant that supports the Model Context Protocol (MCP) using a server address and an OAuth sign-in, and approve what it may do on a consent screen. Access is split into three scopes: read, manage and send. There is no API key to copy.

The connect guide has the steps for Claude Code, Codex, Cursor, VS Code and other MCP clients.

What it is built around

An agent's reach is bounded

Anything that could reach more than one person is previewed first: the assistant sees exactly who would receive a message, and the send is only allowed for that exact list and text.

Every automated action is answerable afterwards

Every call an assistant makes, including the ones that were refused, is written to an activity log you can read. If it is not in the log, it did not happen as far as you can verify.

An opt-out is absolute

A contact who opts out is refused identically whether a person in the dashboard or an assistant tries to message them.

What you entered by hand is yours

A sync from WhatsApp fills gaps. It never overwrites something you typed in on purpose.

Your data

Message content is stored on the service to support search, it is stored unencrypted, and it is not used to train AI models. The privacy policy and the terms of service say the rest, including how to have your data deleted.

Contact

Questions, or a request to delete your data: support@messengeragent.dev. You can also read the blog for guides on using an assistant with WhatsApp.