Using LangGraph and MCP Servers to Create My Own Voice Assistant

A personal voice assistant is built to run entirely on a local machine, avoiding API keys and subscription fees, with the initial goal of replicating basic Alexa functionality. The system architecture uses LangGraph for agent logic, Ollama for local LLMs, RealtimeSTT for speech-to-text, and a separate MCP server for smart home device control. A significant part of the project involves selecting an appropriate text-to-speech (TTS) model, with the author evaluating options like Bark, Coqui TTS, and Elevenlabs. The Kokoro TTS model was ultimately chosen for its balance of speed and quality for local execution. The implementation details for the TTS component, including Python class structure and code examples, are provided to demonstrate how it integrates into the overall voice assistant.