music21-mcp
Exposes 16 music21 MIDI analysis and editing tools to AI agents, enabling them to read, edit, generate, and analyze MIDI files through natural language.
README
music21-mcp
MCP server exposing 16 music21 MIDI analysis and editing tools to AI agents. Built spec-driven with TDD — 120 tests, all green.
What it does
Give an AI agent a MIDI file and it can:
- Read — parse MIDI, detect key, analyze chord progressions
- Edit — transpose, change velocity, filter notes, replace chords, merge files, quantize timings
- Generate — reharmonize progressions, harmonize melodies
- Analyze — detect modulations, extract melody, identify song form, search for patterns
Tools
| Phase | Tool | What it does |
|---|---|---|
| 1a | parse_midi_file |
Load a .mid file into a music21 stream |
| 1a | analyze_key |
Detect key, mode, and confidence |
| 1a | export_midi |
Write a stream to a .mid file |
| 1c | transpose_midi |
Transpose by semitones or to a target key |
| 1c | change_velocity |
Scale, set, or offset note velocities (clamped 1–127) |
| 1c | modify_notes |
Remove/select notes by pitch range, shift octaves |
| 1c | replace_chord_at |
Replace a chord at a specific bar/beat |
| 1c | merge_midi |
Overlay two streams (flat or multi-part score) |
| 1c | quantize_midi |
Snap note timings to quarter/eighth/sixteenth grid |
| 2 | reharmonize |
Replace chords with diatonic or jazz substitutions |
| 2 | harmonize_melody |
Generate block chord accompaniment from a melody |
| 3 | analyze_chord_progression |
Identify chords + Roman numerals in a progression |
| 3 | extract_melody |
Isolate the melody line (highest pitch per offset) |
| 3 | detect_modulations |
Find key changes via sliding window analysis |
| 3 | analyze_form |
Identify song sections (A, B, A') by register changes |
| 3 | search_pattern |
Find melodic patterns by exact pitch or interval sequence |
Getting started
Prerequisites
- Python 3.11+
- A virtual environment (recommended)
Install
cd music21-mcp
# Create and activate venv
python3 -m venv venv
source venv/bin/activate
# Install the package in editable mode
pip install -e .
# Install MCP server dependency
pip install mcp
Run tests
# All 120 tests
pytest -v
# Just one tool's tests
pytest tests/test_transpose_midi.py -v
Use as MCP server
The server runs over stdio, compatible with any MCP-aware client (Claude Desktop, Cursor, etc).
# Start the server
python -m music21_mcp.server
To configure in an MCP client, add to your client's MCP config:
{
"mcpServers": {
"music21-mcp": {
"command": "python",
"args": ["-m", "music21_mcp.server"],
"cwd": "/path/to/music21-mcp"
}
}
}
Or if using the project venv:
{
"mcpServers": {
"music21-mcp": {
"command": "/path/to/music21-mcp/venv/bin/python",
"args": ["-m", "music21_mcp.server"]
}
}
}
Use directly in Python
Every tool is a pure function you can call without the MCP layer:
from music21_mcp.tools.parse_midi import parse_midi_file
from music21_mcp.tools.analyze_key import analyze_key
from music21_mcp.tools.transpose_midi import transpose_midi
from music21_mcp.tools.export_midi import export_midi
# Load a MIDI file
stream = parse_midi_file("song.mid")
# Detect the key
key_info = analyze_key(stream)
print(key_info)
# {'key': 'C major', 'mode': 'major', 'confidence': 0.797}
# Transpose to F major
transposed = transpose_midi(stream, target_key="F")
# Save the result
export_midi(transposed, "song_in_f.mid")
Example: the full pipeline
from music21_mcp.tools import (
parse_midi_file, analyze_key, transpose_midi,
change_velocity, quantize_midi, export_midi,
)
# Load
s = parse_midi_file("input.mid")
# Detect key
print(analyze_key(s))
# Transpose up 2 semitones, make 20% louder, quantize to 16th notes
s = transpose_midi(s, semitones=2)
s = change_velocity(s, scale=1.2)
s = quantize_midi(s, grid="sixteenth")
# Save
export_midi(s, "output.mid")
Example prompts for testing
Single-tool prompts to verify each tool works in isolation:
-
Parse only — "Load
song.midwith parse_midi_file and tell me how many parts and measures it has." -
Key detection — "Use parse_midi_file to load
song.mid, then analyze_key to tell me the key, mode, and confidence." -
Export — "Load
song.midwith parse_midi_file and export it tocopy.midwith export_midi. Did the file size match?" -
Transpose — "Parse
song.midand transpose it down 3 semitones. Export tosong_down.mid." -
Transpose to key — "Load
song.mid, detect its key, then transpose it to E minor. Export tosong_em.mid." -
Velocity — "Parse
song.mid, set all velocities to a fixed value of 90. Export tosong_vel90.mid." -
Velocity boost — "Load
song.midand scale all velocities by 1.3 (clamped to 127). Export tosong_louder.mid." -
Filter notes — "Parse
song.midand remove all notes below C3. Export tosong_no_bass.mid." -
Octave shift — "Load
song.midand shift everything up 1 octave. Export tosong_oct_up.mid." -
Replace chord — "Parse
song.midand replace the chord at bar 2, beat 1 with a C major triad. Export tosong_newchord.mid." -
Merge — "Load
drums.midandbass.mid, merge them into a single file. Export todrum_and_bass.mid." -
Quantize — "Parse
song.midand quantize all timings to a sixteenth-note grid. Export tosong_quantized.mid." -
Reharmonize — "Load
song.midand reharmonize the chord progression using jazz substitutions. Export tosong_jazz.mid." -
Harmonize — "Parse
melody.midand generate block chord accompaniment from the melody line. Export tomelody_with_chords.mid." -
Chord analysis — "Load
song.midand analyze the chord progression. Give me the chords and Roman numerals." -
Extract melody — "Parse
song.midand extract just the melody line. Export tomelody_only.mid." -
Detect modulations — "Load
song.midand find all key changes. Tell me where each modulation happens." -
Analyze form — "Parse
song.midand identify the song structure (A, B, A' sections)." -
Search pattern — "Load
song.midand search for the pitch sequence C4-E4-G4. Tell me where it occurs."
Multi-tool chains to test tool composition:
-
Full analysis — "Load
song.mid. Detect the key, analyze the chord progression, extract the melody, and analyze the song form. Give me a complete analysis report." -
Transpose and harmonize — "Parse
song.mid, transpose up 5 semitones, then reharmonize the result with jazz substitutions. Export tosong_jazz_up.mid." -
Clean and quantize — "Load
song.mid, remove all notes below C2, quantize to an eighth-note grid, and scale velocities by 0.8. Export tosong_cleaned.mid." -
Melody extraction pipeline — "Parse
song.mid, extract the melody, harmonize it with block chords, and export tomelody_harmonized.mid." -
Key-aware reharm — "Load
song.mid, detect the key, then reharmonize using diatonic substitutions appropriate for that key. Export tosong_diatonic.mid." -
Modulation-aware transpose — "Parse
song.mid, detect all modulations, then transpose the whole piece so the first key becomes A minor. Export tosong_am.mid." -
Pattern search then replace — "Load
song.mid, search for the interval sequence [2, 2] (two whole steps up), and replace each match with a diminished triad. Export tosong_dim.mid." -
Form-aware analysis — "Parse
song.mid, analyze the form to find the B section, then extract just the melody from that section and harmonize it. Export tob_section_harmonized.mid." -
Multi-file merge and clean — "Load
drums.mid,bass.mid, andkeys.mid. Merge all three, quantize to a sixteenth grid, and scale velocities to 100. Export tofull_band.mid." -
End-to-end remix — "Load
song.mid. Detect key, transpose down 2 semitones, quantize to eighth notes, reharmonize with jazz substitutions, boost velocity by 1.4, and export tosong_remix.mid."
Architecture
src/music21_mcp/
__init__.py # Package entry
server.py # FastMCP server — 16 tool wrappers returning JSON
tools/
parse_midi.py # Phase 1a
analyze_key.py
export_midi.py
transpose_midi.py # Phase 1c
change_velocity.py
modify_notes.py
replace_chord_at.py
merge_midi.py
quantize_midi.py
reharmonize.py # Phase 2
harmonize_melody.py
analyze_chord_progression.py # Phase 3
extract_melody.py
detect_modulations.py
analyze_form.py
search_pattern.py
tests/
test_*.py # One test file per tool, plus server + e2e tests
Each tool is a pure function that takes a music21 Stream and returns a Stream or dict. The MCP server layer handles file-path-to-stream conversion and JSON serialization. This separation means tools are testable without the MCP protocol and usable as a plain Python library.
Tech
- music21 — MIT's music theory and analysis toolkit
- MCP — Model Context Protocol for AI agent tool calling
- Python 3.11+, pytest, strict TDD
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.