Dev.to AI 🤖 Ai 👁 0 📖 2 min read

Give your AI agent WhatsApp, SMS, and email with Zavu's MCP server

AI agents are getting good at reasoning. They are still awkward at talking to people on the channels those people actually use. If your agent lives in Cursor, Claude, or a terminal, it usually cannot send a WhatsApp mess

AI agents are getting good at reasoning. They are still awkward at talking to people on the channels those people actually use. If your agent lives in Cursor, Claude, or a terminal, it usually cannot send a WhatsApp message, an SMS, an email, or answer a phone call without a pile of custom integrations.

Zavu is communication infrastructure for AI agents. One API, one webhook contract, and one bill cover WhatsApp, SMS, voice, email, Telegram, Instagram, and Messenger. You can send and receive on those channels, or deploy an agent that answers them. There is a free plan with no credit card: 2,000 messages a month on WhatsApp, Telegram, Instagram, and Messenger, plus a small credit for SMS, voice, and email.

This post is about the fastest path I have found: the MCP server, so a coding agent can call Zavu as a tool.

Install the MCP server

The local server is an npm package. It needs a Zavu API key from the dashboard:

npx -y @zavudev/sdk-mcp

Set ZAVUDEV_API_KEY in the environment. Docs live at docs.zavu.dev/tools/mcp-server.

There is also a remote, read-only MCP endpoint at https://docs.zavu.dev/mcp. That one searches the documentation and collects feedback. It does not send messages.

Add the agent skills

If you use a coding agent that supports skills, you can install Zavu's public skill pack:

npx skills add zavudev/zavu-skills

The repo is github.com/zavudev/zavu-skills (Apache-2.0). The skills walk the agent through login, sending a message, and deploying.

Send a first message from the CLI

You do not have to start with MCP. The CLI is enough to see a message go out:

npx zavudev@latest login
npx zavudev messages send --to +14155551234 --text "Your order #1234 has shipped."

Replace the number with a recipient you are allowed to message.

Deploy an agent on those same channels

Zavu is not only a messaging pipe. You can declare an agent in TypeScript with defineAgent and defineTool, then run:

npx zavudev deploy

The agent answers on the same channels, including voice calls. Official SDKs cover TypeScript, Python, Ruby, Go, and PHP. The OpenAPI spec is at docs.zavu.dev/openapi.json.

WhatsApp itself is billed by Meta at cost, with no markup from Zavu. Other channels use the free allowance or prepaid credit on the pricing page.

Why this belongs in an agent stack

Most CPaaS tools were built for apps, not for agents. You still end up writing a WhatsApp integration, then SMS, then email, each with its own auth and webhook shape. Zavu gives agents a single tool surface: send, receive, route, and, if you want, run the agent itself.

If you are wiring an agent that should talk to customers, start at zavu.dev and the MCP server docs.

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.