mohitbadwal/ringback

What it does

Ringback is an MCP (Model Context Protocol) server that enables large language models to initiate and conduct live voice calls directly. It bridges LLM capabilities with real-time voice communication, allowing the AI to reach users on their phones with interactive, interruptible conversations. The tool supports tiered alert systems and maintains conversation context throughout the call.

How to set up

Ringback is self-hosted and requires: (1) SIP infrastructure via pjsua2 or Linphone for voice call handling, (2) Whisper.cpp for speech-to-text conversion, and (3) macOS say or equivalent TTS for text-to-speech synthesis. Installation involves cloning the repository, configuring your SIP credentials, and running the MCP server to connect with your LLM platform.

Key features

• Live, interruptible voice calls initiated by LLMs
• Self-hosted SIP telephony with no paid service dependencies
• Whisper.cpp-based speech recognition for cost-free transcription
• Native macOS text-to-speech synthesis
• Tiered alerting system for priority-based notifications
• No additional API keys required for conversation
• Open-source and fully customizable