The idea behind MXP4 was simple. Every recorded song should be an instrument you can play. Not a fixed waveform you press play on, but a packaged set of stems and metadata that lets you mute the drums, isolate the vocals, build a different version, sing along, and share what you made. In 2008, three French entrepreneurs founded a Paris startup around that thesis. They built a file format, an in-browser player, and a business that for a brief moment included a licensing deal with the Jackson family for the Jackson 5 catalogue.

It lasted five years.

The story of MXP4 is the story of a company that got the product right and the moment wrong. Eighteen years later, almost every piece of the technology landscape they were missing has fallen into place. Anyone building remix tools in 2026 is, in a meaningful sense, finishing what they started.

The founders and the format

MXP4 was founded by Gilles Babinet, Sylvain Huet, and Philippe Ulrich, a trio with a serious music-and-tech background. Babinet later became the French government's Digital Champion at the European Commission. Ulrich was already a known figure in the French digital arts scene from his work at Cryo Interactive. Huet was the technical lead.

The format they designed, called MXP4 like the company, packaged everything an interactive listener might want into a single .mxp4 file. Streaming music tracks, separated stems, video, artist biography, concert listings, social share buttons, links to buy merchandise. The container was opinionated. The promise was that you could open a .mxp4 in their web player and the song would behave less like a recording and more like an app.

Inside the player, listeners could strip away stems, recombine them, sing along with vocal isolation, build personal mixes, and share the result back through the same socially-aware container. In 2009 the company struck a partnership with the Jackson estate to package Jackson 5 catalogue tracks as interactive .mxp4 files. The launch coverage was real. The product worked.

What they got right

Almost the entire high-level thesis is what Remix.me, Moises, Splitter, and the rest of the 2026 stems-and-remix industry is now built on.

  • Stems as the primary unit. Not the mixdown. Not the EQ band. Not the genre. The individual recorded tracks that make up a song. MXP4's container exposed them directly.
  • Interactivity as the experience. The job of the player was to let the user do something with the song, not just press play.
  • Social as the distribution. Shareable, embeddable, friend to friend. A remix that does not get shared is a remix nobody sees.
  • Major-label licensing as the strategy. Not unlicensed UGC. Not orphan catalogue. Real songs from real artists, packaged with real rights.
  • Format-first thinking. Define the container. Let others build on it.

Each of those decisions reads in 2026 like a starting assumption you would put into the design document for a remix app today.

What killed it

The product was right. The conditions were wrong, in five overlapping ways.

Flash and the death of browser plugins. The MXP4 player ran on Adobe Flash. The Flash deprecation timeline (Apple dropping it on iOS in 2010, the long retreat through the 2010s, the final shutdown in 2020) meant that the entire delivery surface for MXP4 was a sinking ship from launch day. The format was designed for a web that was already being replaced.

Mobile as the music platform. The iPhone was a year old when MXP4 launched. The App Store did not open until July 2008. The entire music industry was about to migrate to mobile-first listening, and MXP4 was a desktop-browser-plugin experience. By the time mobile listening dominated, MXP4 had no native story.

Server-side stem economics. Stems in 2008 had to be delivered by the rights holder, as files, mixed and mastered specifically for the format. There was no on-device stem separation. No AI model existed that could decompose a master into its tracks. Every .mxp4 file was a custom production project. That meant the catalogue was always going to be tiny relative to the universe of recorded music.

Major-label licensing as a treadmill. The Jackson 5 deal was a coup, but every additional artist or label required a new bespoke negotiation, new clearances, new minimum guarantees. The unit economics never caught up to the speed of consumer adoption. MXP4 stayed catalogue-constrained the entire time it existed.

No business model that scaled. Was MXP4 a B2B technology licensor? A consumer destination? A social network? A label tool? A games company? The 2010 pivot to Bopler, a Facebook-era music-tapping game brand, suggests the team was looking hard for any wedge that would close the loop between the technology and a payday. The wedge did not appear in time. The company became inactive around 2013.

What is different in 2026

MXP4 is worth thinking about in 2026 because each of those five killers is now solved.

Flash is gone, and HTML5 plus native apps replaced it cleanly. An interactive music player no longer needs a browser plugin. It can be a desktop app, a phone app, an AUv3 extension inside any DAW. Distribution is a solved problem.

Mobile-first is the default. Every device that matters runs apps. Streaming subscriptions are universal. The audience MXP4 was reaching for, fans who want to interact with music rather than just consume it, now lives entirely on platforms that support real interactive playback.

On-device stem separation works. Demucs, BS-Roformer, AudioShake's commercial models. The AI separation quality crossed the "actually usable" threshold around 2022 and has been improving since. On an Apple Silicon Mac, a four-minute track separates into clean stems in under thirty seconds, locally, with no upload. The server-side stem economics that strangled MXP4's catalogue have been replaced by client-side processing of any track the user has streaming rights to.

Streaming-service partnerships are a known pattern. Beatport Streaming, Tidal, SoundCloud Go+, and the long-running djay Pro / Apple Music relationship have established that authenticated subscribers can have their streams interactively processed inside third-party apps. The licensing model has precedent. The legal envelope is the user's own subscription, not a label deal.

Per-derivative licensing is becoming a real category. The Universal Music Group and Spotify agreement announced in May 2026, covering AI-generated covers and remixes inside Spotify Premium with a per-stream royalty flow, is the first major-label deal explicitly designed for derivative-work compensation. It is Spotify-exclusive in-platform and does not help non-Spotify tools directly, but it signals that the major labels have, after eighteen years of MXP4-shaped objections, made peace with the idea of licensing derivatives at scale.

The lessons

If you are building remix tools in 2026, MXP4 leaves you four things worth taking seriously.

One. Get the format-vs-instance distinction right. MXP4 was building containers full of audio. The modern equivalent is a recipe: a small JSON file with references to stems the user already has rights to, plus their own additions. The audio does not travel. The recipe does. That sidesteps catalogue-as-product economics entirely.

Two. Do not build on a sinking distribution surface. MXP4 picked Flash. A 2026 remix tool that depends on a single streaming partner's continued goodwill, or a single mobile OS, or a generative-AI moment that may or may not last, is making the same bet in a different costume. Diversify the surface.

Three. Let the user bring the rights. The cleanest legal posture for any remix tool is the one where the user is the rights holder, through their own streaming subscription, their own local files, their own purchased catalogue. The platform's job is to make remix creation easy on top of those rights, not to acquire and resell catalogue.

Four. Have a monetization story before the catalogue story. MXP4 chased catalogue while still looking for a business model. The order should be reversed. If the product makes money from individual users (a freemium app, a subscription, a per-export fee), the catalogue conversation with rights holders becomes a "we already have an audience" conversation, not a "we need your audience" one.

The bigger picture

MXP4 existed in a window where the music industry was still arguing about whether streaming should exist, where the smartphone was still proving itself, where Flash was the web. The startup was trying to skip ahead by two technology generations. They did not have the AI to separate stems on demand. They did not have the distribution surface to reach a global audience. And they did not have a streaming-services ecosystem willing to let third parties touch licensed audio.

Eighteen years later, all three of those exist. The product idea, every recorded song as an instrument you can play, is in 2026 mostly an execution problem.

That is the kind of problem startups are supposed to solve.