WACrew

Guide

MCP (Model Context Protocol), explained through WhatsApp

MCP is an open protocol that lets an AI agent use external tools. With a WhatsApp MCP server, Claude Code, Codex or Cursor can read and answer your conversations.

Updated 1 min read
On this page
  1. Where you meet it
  2. Common mistake
  3. Questions

MCP (Model Context Protocol) is an open protocol, released by Anthropic in 2024, that standardizes how an AI agent uses external tools and data. An MCP server exposes tools ("read messages", "transcribe audio", "reply"); an MCP client (Claude Code, Codex, Cursor, Gemini CLI, Antigravity and others) offers them to the model, which decides when to call each one.

Where you meet it

You register the server in your client, usually with one command or a JSON config, and then ask in plain language: "who is waiting for a reply?", "summarize the project X group", "transcribe the voice note he sent". The agent picks the tool, gets the result and answers.

WACrew's MCP server (wacrew-mcp, on npm and in the official MCP Registry) has 37 tools over the collector's history: reading, search, media, triage, calendar, voice and sending at a human pace. Setup in WhatsApp MCP server.

Common mistake

Thinking MCP is "an AI". It is not: it is the plug between the AI and the tool. Without a server the agent cannot see your WhatsApp; without a client the server just sits there.

Another one: handing the agent a send tool with no rules. In WACrew every chat starts in confirm mode, where the agent shows the text and waits for your OK.

Questions

Is MCP Anthropic-only? The protocol is open and used by clients from several companies.

Does MCP work with a local model? Yes, if the client you run the local model in supports MCP.

What is the difference between MCP and an API? The API is the service; MCP is the format that describes its tools so an agent can use them.

More in what is MCP. In Portuguese: MCP.