Viberadio Fm MCP server

VibeRadio

Your agent's output as audio.

Turns an agentic coding session into something you can listen to instead of read. Narration is the core — spoken monologue, two-host podcast, or reactive music — so you can follow a Claude Code / Codex / Gemini agent's progress while away from the desk, commuting, or multitasking. Reactive music (below) is one style among several.

Styles: spoken monologue · two-host podcast · reactive music (lo-fi, synthwave, ambient, classical, 8-bit). Modes: ambient stream (narrates live as the agent works) · task summary (one message when a task finishes) · session recap (a longer podcast/musical wrap). Timing: sync/live (updates as the agent works) or async (after a turn/session) — user-configurable, per-hook. See spec.md.

Demo

▶ Watch the launch video — four coding agents, four radios, each tuned to its own station.

https://github.com/pooriaarab/viberadio/raw/main/branding/launch-video.mp4

Quick start

npm i -g viberadio-fm

viberadio say "hello"                       # narrate a line now (on-device, offline)
viberadio recap session.json                 # build a script from session events and speak it
viberadio recap session.json --style podcast  # two-host podcast style

Works offline, on-device, with zero keys. v0 routes all narration through the on-device tier of the shared @pooriaarab/vibe-core model cascade — macOS say / Linux espeak / spd-say — so it runs with nothing configured and nothing leaves your machine. Bring-your-own-key TTS (ElevenLabs, OpenAI, …) plugs into the same cascade later; the createNarrator() seam is already there.

Use it as a library too:

import { narrate, recap, buildRecapScript } from 'viberadio-fm';

await narrate('build passed');                              // speaks via on-device TTS
const { script } = await recap(events, { style: 'podcast' }); // script + speak a session
buildRecapScript(events);                                   // pure: events -> script, no audio

MCP

viberadio mcp starts a stdio MCP server exposing narrate({ text, style? }) and recap({ events, style?, mode? }) tools, so an agent can ask for an audio recap mid-session. Add it to your client's MCP config:

{
  "mcpServers": {
    "viberadio": { "command": "viberadio", "args": ["mcp"] }
  }
}

Why Build This

• Developers already listen to lo-fi while coding — this makes it reactive and personal
• Extremely visual and shareable — imagine a tweet: "My code literally makes music"
• No one has done reactive music for coding before — genuinely novel
• Combines two massive communities: developers and lo-fi/music lovers
• The demo video alone could go viral — watching music change as code is written

Features

• Reactive soundtrack — Music dynamically shifts based on coding events
• Commit → beat drop
• Test pass → melodic resolution
• Test fail → minor key shift
• Error/exception → dramatic tension
• Idle → ambient drone
• Deploy → triumphant crescendo
• Genre modes — Lo-fi, synthwave, ambient, classical, 8-bit
• Visualizer — Terminal-based audio visualizer that shows the music reacting
• Session recording — Save your coding soundtrack as an MP3/WAV
• Spotify-style wrap — Weekly "Your Vibe Radio Wrapped" with stats and generated playlist
• Collaborative mode — Team coding sessions merge into one shared soundtrack

Distribution

• CLI — viberadio start — runs in background, hooks into git/terminal events
• npm package — npm install -g viberadio-fm
• Claude Code hook — Auto-starts when Claude Code session begins
• Claude Code skill — /viberadio to control playback
• skills.sh — Listed on skills.sh marketplace
• macOS menubar app — Visual controls + now-playing display

Shortened. The full README is on GitHub.