# Aurora MCP Server [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/EricDisero/aurora-mcp  
**GitHub Stars:** 2  
**npm Downloads (last month):** 751  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/ericdisero-aurora-mcp-server

## Description
MCP server for Aurora, the AI audio workbench. Your AI agent (Claude, Cursor, or any MCP client) drives music generation, covers, sample manufacturing, 7-stem separation, and DAW-ready stem organization. Files land on disk.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "aurora-mcp-server": {
    "command": "npx",
    "args": ["-y","@ericdisero/aurora-mcp-server"],
    "env": {
      "SUNO_API_KEY": "",
      "KIE_API_KEY": "",
      "MVSEP_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `SUNO_API_KEY`, `KIE_API_KEY`, `MVSEP_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Aurora MCP Server does

Aurora MCP Server MCP server gives AI agents access to the Aurora AI audio workbench. It supports music generation, covers, sounds, vocal and instrumental layers, section edits, extensions, and mashups through Suno. Agents can also create projects and tracks, import audio or references, move assets, mark favorites, and inspect recipes used by existing assets.

The server handles audio separation and organization as well. It can split an asset into vocals, kick, snare, toms, hats, bass, and Other, or extract instrument groups, individual instruments, and vocal modes. Generated and processed files are placed in project folders that the Aurora desktop app shares through SQLite and the local filesystem. The desktop app does not need to be running.

## How it works

The Aurora MCP Server MCP server is distributed as a Node.js package using stdio. Installable packages provide the MCP server, the `aurora` CLI, and shared functionality. The complete operation list is available through MCP `tools/list` or `aurora run --list`; version 0.6.2 documents 55 tools and 5 skills.

Suno generation and MVSEP separation use provider credentials and may consume credits or premium minutes. Separation can first be planned with `estimateOnly: true`, which reports routes and expected units without submitting work. Started split and extraction jobs return a job ID and progress through queued, submitting, waiting, and landing states. Status calls can advance work, poll providers, and save checked files. Job manifests persist under the Aurora user-data directory, allowing work to resume after restarts.

Provider separation results undergo local checks before stems are saved. Checks validate output keys, audio shape, samples, route sums, and applicable family tests. A passing check does not prove that every musical label is perfectly separated.

## Setup and configuration

Node.js 18 or newer is required. Install the CLI with:

```bash
npm i -g @ericdisero/aurora-cli
```

Configure `SUNO_API_KEY` or its `KIE_API_KEY` fallback for Suno operations, and `MVSEP_API_KEY` for provider separation. Credentials can be supplied in the MCP client environment or stored with `aurora keys`; saved settings go to `~/.aurora/config.json`, while environment variables take precedence.

A stdio client can launch `npx -y @ericdisero/aurora-mcp-server`. A local checkout can instead run its built `packages/mcp/dist/server.js` with Node. Local library reads, route discovery, checks, job snapshots, cancellation, and local audio processing do not require provider keys.

## Tools and capabilities

The tool surface includes capabilities for:

- Generating tracks, covers, sounds, layers, extensions, replacements, and mashups.
- Creating and editing local projects, tracks, assets, references, and favorites.
- Listing separation routes and planning split or extraction work before submission.
- Starting, monitoring, checking, resuming, listing, and cancelling jobs.
- Reading saved split, extracted, imported, and custom stem sets.
- Converting or pitch-shifting files locally.
- Measuring beat grids from audio with the local `aurora_beat_grid` tool.

The beat-grid operation accepts an asset ID or absolute path and uses the bundled Beat This! environment after one-time Python setup. RVC and MIDI sidecars require the Aurora repository and their Python dependencies.

## Limitations and notes

Mix, mastering, and Export are not exposed as agent-controlled desktop operations. The server does not provide a bridge for driving those live app flows.

Suno generation, covers, layering, editing, and provider WAV conversion can spend credits. MVSEP submissions consume premium minutes per planned call. Queuing a job does not itself submit provider work, but advancing status can do so. Submitted provider work may continue after cancellation and is not refunded.

Downloaded preview URLs from Suno expire; downloaded files are the durable outputs. Partial jobs retain successful files and diagnostics. Inspect fields such as `lastError`, `callResults`, and `extractedFiles` before retrying. Local checks are free, but some instrument-family label swaps can pass the available identity and sum checks.

_Full upstream README: https://allmcps.com/mcp/ericdisero-aurora-mcp-server/readme_