MCP server for music analysis: connect Claude, Cursor or ChatGPT to a song

October 17, 2026 · 6 min read

What the server can do

The server exposes seven tools. The first four answer without any key, so an assistant can show you a real response on the first try. Registry name: io.github.songbrain-ai/songbrain in the official MCP Registry. The older path /mcp still answers too.

Check it yourself without a client:

curl -X POST https://api.songbrain.ai/v1/mcp \
  -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Connect it

Leave the header out to use the example tools only. Add it with a key from the Developer Console to analyse your own songs.

claude mcp add --transport http songbrain https://api.songbrain.ai/v1/mcp \
  --header "Authorization: Bearer sb_live_YOUR_KEY"

https://api.songbrain.ai/v1/mcp

{
  "mcpServers": {
    "songbrain": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.songbrain.ai/v1/mcp",
               "--header", "Authorization: Bearer sb_live_YOUR_KEY"]
    }
  }
}

{
  "mcpServers": {
    "songbrain": {
      "url": "https://api.songbrain.ai/v1/mcp",
      "headers": { "Authorization": "Bearer sb_live_YOUR_KEY" }
    }
  }
}

https://api.songbrain.ai/v1/mcp

Then ask: "Open the Songbrain example 'the-fallen-heroes', tell me where the chorus starts and which 15 seconds you'd use for a Reel."

What the assistant gets back

One document per song, times in seconds from the start. The summary view fits in a chat; view="full" adds every beat and every word timing.

• song_dna: genre and subgenre, tempo, key and key changes, mood, vocals, instruments, energy, LUFS loudness.
• timeline: every beat, every downbeat, time signature, sections with repeat letters (A, B, A, C) and energy.
• best_moments: ranked windows with start, peak and end, and the reason each one works.
• lyrics: lines and words with their own start and end times.
• story + shot_plan: what the song is about, a world and palette, and a beat-synced scene list with a prompt per scene.

The other music MCP servers, honestly

Sources, read on October 17, 2026: hugohow/mcp-music-analysis ("Integrate librosa, whisper with LLMs to analyze music audio", MIT) and gorums/music-mcp-server("Music Collection MCP Server", metadata management and album classification for a local library). Pick the librosa server when you want raw measurements on files that never leave your machine. Pick the Music Collection MCP to tidy a library. Pick Songbrain when the assistant should understand the song: where the chorus is, which part to use, what is sung, what to show.

Good prompts to start with

• "List the Songbrain example songs and open the one closest to my genre."
• "Analyse this song: https://…/track.mp3. When it's done, give me the section map and the best 15 seconds with the lyric sung there."
• "Take the shot plan and rewrite the first three scenes as Kling prompts in a neon-noir style."
• "Compare the energy per section with the lyrics and tell me where the song lifts and where it drags."