GitHub Repository

Teach your AI to play piano and guitar — and sing. 46 MCP tools, 120 fully annotated songs, 6 engines, a live browser cockpit. Ships jam-actions-v0, a public dataset of MCP tool-use traces over classical piano.

3 starsTypeScript

AI Jam Sessions – MCP server that teaches AI to practice piano

Name: AI Jam Sessions – MCP server that teaches AI to practice piano
Availability: InStock
Author: mikeyfrilot

by mikeyfrilot·Feb 24, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerZero to OneWizardryBig Brain

Teaches LLMs to play piano by giving them ears, eyes, and persistent practice journals.

Strengths

•Genuinely novel framing: embodied music learning for LLMs (reading, hearing, seeing, remembering, singing).
•120 annotated MIDI songs + six audio engines (oscillator, sample, vocal tract, additive, guitar) is comprehensive.
•Practice journal persistence across sessions compounds learning—real pedagogical insight applied to AI.

Weaknesses

•Early-stage MCP ecosystem—will live or die based on Claude/LLM adoption of MCP servers.
•No evidence of actual LLM learning outcomes; framing is clever but empirical results undemonstrated.

Post Description

Built this as an experiment in giving LLMs embodied musical experience. It's an MCP server with 120 MIDI songs across 12 genres. Each genre has one fully annotated exemplar the model studies first. The rest are raw MIDI waiting to be learned.

The model can read sheet music, play songs through your speakers, view an SVG piano roll it can read back to verify what it played, and write to a practice journal that persists across sessions. Learning compounds over time.