This MCP (Model Context Protocol) server integrates MuseScore with LLM clients like Claude Desktop, enabling basic music composition through natural language.

Brief demo here: https://www.canva.com/design/DAGlHeR706g/4UcXvK23I1axWOfm6lSSxQ/edit

The server allows an MCP-connected LLM to:

• Connect to MuseScore and manipulate an open score
• Add notes and rests
• Select and delete
• Create tuplets
• Undo a change
• Move from staff to staff

• Can not understand multiple voices within a staff or navigate between them

• Python 3.9+
• MuseScore 3 or 4 installed
• MCP Python SDK

1. Clone this repository:

   git clone https://github.com/yourusername/musescore-mcp-server.git
   cd musescore-mcp-server
   

2. Install dependencies, either in a venv or globally:

   pip install -r requirements.txt
   

3. Install the musescore-mcp-plugin by copying it into your MuseScore plugins dir (on my mac this was ~/Documents/MuseScore4/plugins)

4. Configure your LLM with the MCP server (server.py). If you are using Claude Desktop and a venv, this can be done by creating a claude_desktop_config file like so:

   {
    "mcpServers": {
      "musescore": {
        "command": "bash",
        "args": [
          "-c",
          "source /path/to/venv/activate && python3 /path/to/server.py"
        ]
      }
    }

Open musescore and whatever score you want to use. Then, connect the musescore-mcp-plugin and launch it by selecting it from the plugins menu.

Once connected, you can ask Claude questions like:

• "Help me come up with some possible chords for the melody in my score
• Compose a simple melody and show me 4 ways to harmonize it
• Extend the melody in measures 1-12 with 4 measures that match the style of this piece

If you want to add features to the MuseScore plugin and test them, the testClient.html file in this repo may be a useful development tool - simply extend it to call whatever functions you add.