🤖 WhatsApp Agent (MCP + Google ADK)

An intelligent WhatsApp Agent built with Google ADK and the Model Context Protocol (MCP).
Send messages, search chats, and interact with WhatsApp directly from your AI agent environment.

---

✨ Features

• 📩 Send WhatsApp messages (with AI-crafted text + disclaimers).
• 🔍 Search chats & messages using MCP tools.
• 🧩 Schema sanitization for Gemini compatibility.
• ⚡ Powered by Google ADK, MCP Toolsets, and WhatsApp bridge.
• ☁️ Deployment-ready with environment configs.

---

🧩 Architecture

+-------------------+        +-----------------+        +---------------------+
|   User / Agent    | <----> |   MCP Toolset   | <----> | WhatsApp MCP Server |
+-------------------+        +-----------------+        +---------------------+
                                   (Python)                   (Go + SQLite)

• Go Bridge → connects to WhatsApp multi-device API.
• Python MCP Server → exposes standardized tools (send, search, list).
• Google ADK Agent → orchestrates message crafting + disclaimer + sending.
• WhatsApp MCP Server (repo) → lharries/whatsapp-mcp

---

⚙️ Installation

Prerequisites

• Go
• Python 3.9+
• UV (Python package manager)
• WhatsApp mobile app (for QR authentication)

Setup

1. Clone this repository

   git clone https://github.com/B4K2/WhatsApp-agent.git
   cd WhatsApp-agent

2. Install dependencies

   uv sync

3. Set environment variables (in .env)

   UV_EXECUTABLE_PATH=/usr/local/bin/uv
   PYTHON_MCP_SERVER_DIRECTORY=./multi_tool_agent

4. Run the WhatsApp MCP bridge

   cd whatsapp-bridge
   go run main.go

   📲 Scan the QR code with WhatsApp mobile to authenticate.

5. Run the Python MCP server

   uv run main.py

---

🚀 Usage

Here’s how the agent works:

from google.adk.agents import Agent

agent = Agent(
    name="WhatsApp_Orchestrator",
    model="gemini-2.0-flash",
)

# Example: Send a message
agent("Send 'On my way!' to +91XXXXXXXXXX")

✅ Agent crafts the message ✅ Appends disclaimer ✅ Sends via send_message tool

---

⚠️ Security Note

As with many MCP servers, be cautious of prompt injection attacks. Messages and metadata are stored locally in SQLite.

---

🤝 Contributing

Pull requests are welcome! If you’d like to extend the toolset (media sending, chat search), fork and submit a PR.