Voice control for AI coding agents. Built with LiveKit and Solo.

I built a voice bridge that lets me reach my coding agents like Claude Code, Codex, and Gemini from anywhere with cell service. Call a phone number or open a web app, and talk to Claude Code, Codex, and Gemini CLI running on my laptop while they are still working. Status checks, unblocking stuck agents, and answering yes-or-no prompts, all by voice.

In this video I walk through the architecture I landed on: wrapping Solo's local MCP server with a LiveKit voice agent, narrowing 78 MCP tools down to roughly 10 voice-safe ones, and why this voice interface for the terminal belongs natively inside Solo.

What you will learn:
- How to wrap an MCP server with a voice agent
- The peer-bus model that makes a voice agent for developer tools possible
- Phone access using LiveKit Phone Numbers and a SIP dispatch rule
- Web access using a Next.js client and a LiveKit token route
- Curating a large MCP tool surface down to a voice-safe subset
- The supervisor pattern as the natural next step
- How LiveKit's free tier covers a single-user setup

Tools and references:
- Solo: https://soloterm.com
- LiveKit Agents SDK: https://docs.livekit.io/agents/start/voice-ai/
_____________________________________

👀 Watch Next:
Playlist: Web Development For Beginners - https://www.youtube.com/watch?v=Ez4yHS2dsN8&list=PLkwxH9e_vrAJ0WbEsFA9W3I1W-g_BTsbt
_____________________________________

🌐 Connect With Me:
Website: https://www.codestackr.com
Twitter: https://twitter.com/codeSTACKr
Instagram: https://instagram.com/codeSTACKr
Facebook: https://facebook.com/codeSTACKr
_____________________________________

** Affiliate Disclaimer: Some of the above links may be affiliate links, which may generate me a sales commission at no additional cost to you.

#claudecode #aiagents #vibecoding

codeSTACKr

Solo is a desktop workspace that runs multiple AI coding agents (Claude Code, Codex, Gemini) side-by-side, with each pane communicating via a local MCP server bus. The author built an external bridge using LiveKit to enable remote access to Solo's agent processes via phone call or a Next.js web app. This lets developers check on stuck agents, send voice or text responses, and keep work moving when away from their laptop. The architecture uses LiveKit for telephony, WebRTC, and voice loop, while a lightweight voice agent connects to Solo's MCP tools over stdio. The tool set is narrowed from 78 to ~10 tools to keep the voice LLM manageable, with a custom fan-out tool for cross-project queries.

Claude Code, Codex, and Gemini in ONE app