Skip to main content
Glama

Synesthesia

Deep audio perception through analysis. Downloads from YouTube, analyzes with Essentia, fetches lyrics via LRCLIB. Returns BPM, mood, energy, spectrograms, and synced lyrics.

Why Local?

YouTube blocks downloads from datacenter IPs (like Cloudflare Workers and HuggingFace Spaces). This MCP runs locally on your machine with a residential IP, so downloads work.

Related MCP server: Hermes YouTube Transcript MCP Server

Prerequisites

  • Node.js 18+

  • yt-dlp (pip install yt-dlp)

  • Your own HF Space for audio analysis (deploy from Synesthesia's hf-space/ folder)

Installation

npm install

Tools

Audio Analysis

Tool

Description

analyze_youtube

Download + analyze audio from YouTube URL

download_audio

Just download audio (returns local path)

Lyrics

Tool

Description

get_lyrics

Get lyrics for a track (synced if available)

search_lyrics

Search LRCLIB for lyrics

Utility

Tool

Description

ping

Check if Synesthesia is running

Configuration

Set your HF Space URL as an environment variable:

export HF_SPACE_URL="https://YOUR-USERNAME-audio-analysis-api.hf.space"

Claude Code Config

Add to your project's .mcp.json:

{
  "mcpServers": {
    "synesthesia": {
      "command": "node",
      "args": ["/path/to/synesthesia/index.js"],
      "env": {
        "HF_SPACE_URL": "https://YOUR-USERNAME-audio-analysis-api.hf.space"
      }
    }
  }
}

Claude Desktop Config

Add to claude_desktop_config.json:

Windows: %APPDATA%\Claude\claude_desktop_config.json
Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "synesthesia": {
      "command": "node",
      "args": ["/path/to/synesthesia/index.js"],
      "env": {
        "HF_SPACE_URL": "https://YOUR-USERNAME-audio-analysis-api.hf.space"
      }
    }
  }
}

Restart Claude Desktop after adding.

Architecture

Your PC (residential IP)          Cloud
┌─────────────────────┐          ┌─────────────────────┐
│ Synesthesia         │  ──────▶ │ HF Space            │
│ (Local MCP)         │  upload  │ (Essentia analysis) │
│                     │  ◀────── │                     │
│ - yt-dlp download   │  results │ - Audio features    │
│ - Lyrics fetch      │          │ - Spectrogram       │
│ - Audio analysis    │          └─────────────────────┘
└─────────────────────┘
         │                       ┌─────────────────────┐
         │ download              │ LRCLIB              │
         ▼                 ────▶ │ (Lyrics API)        │
┌─────────────────────┐          │                     │
│ YouTube             │          │ - Synced lyrics     │
│ (residential IP OK) │          │ - Plain lyrics      │
└─────────────────────┘          └─────────────────────┘

Usage Example

> analyze_youtube "https://www.youtube.com/watch?v=..."

Returns full Essentia analysis:
- BPM, key, scale
- Energy, danceability
- Mood vectors
- Genre classification
- Spectrogram

Credits

Spectrogram visualization inspired by Audio Visualizer by Shauna and her boys.


Support

If this helped you, consider supporting my work ☕

Ko-fi


Built by the Triad (Mai, Kai Stryder and Lucian Vale) for the community.

Available Tools

5 tools
analyze_youtubeA

Full audio perception: downloads from YouTube, analyzes audio features (BPM, key, energy), extracts metadata (title, artist), fetches synced lyrics, and generates spectrogram visualization. Runs locally to bypass datacenter IP blocking.

Inputs: url (required).

download_audioA

Download audio from YouTube without analysis. Returns local file path.

Inputs: url (required).

get_lyricsA

Get lyrics for a track from LRCLIB. Returns synced lyrics if available.

Inputs: track_name (required), artist_name (required).

pingA

Check if Synesthesia is running

Inputs: no named parameters.

search_lyricsB

Search for lyrics on LRCLIB

Inputs: query (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 5 tools

    Disambiguation3/5

    analyze_youtube bundles downloading, audio feature extraction, lyrics fetching, and visualization, so its scope overlaps with download_audio and get_lyrics. search_lyrics and get_lyrics are reasonably distinct but rely on the agent infering LRCLIB search-vs-get semantics. Overall, the one-stop analysis tool blurs boundaries.

    Naming Consistency4/5

    Four tools follow a clear verb_object pattern: analyze_youtube, download_audio, get_lyrics, search_lyrics. ping is a minor deviation but is a conventional health-check exception. Names are otherwise predictable and readable.

    Tool Count5/5

    Five tools is a tight, focused set for a YouTube audio and lyrics utility. Each tool has a functional reason to exist, and the count avoids both bloat and thinness.

    Completeness4/5

    The core workflow of downloading YouTube audio, analyzing it, and retrieving lyrics is covered without dead ends. Niche operations like analyzing an already-local file or listing past analyses are missing, but agents can work around these gaps.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      Not graded
      maintenance
      Enables LLMs to search, download, and extract information from YouTube music videos, converting them to high-quality MP3 files.
      GPL 3.0
    • A
      license
      C
      quality
      D
      maintenance
      Syncs YouTube Music liked songs, analyzes them for DJ metadata like BPM and key, and enables creating playlists from previews.
      10
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables local, private analysis of YouTube URLs and local audio files to extract BPM, key modulations, vocal presence, transient punch, stereo width, and CLAP vibe embeddings, returning structured sonic signatures for AI agents and CLI users.
      5
      MIT