Generative music has an attribution problem, and everyone in the industry knows it. When a model produces audio, the connection to rights holders is usually severed at the moment of generation: the output exists, but the questions that the music business runs on — who contributed, who is owed, what can this be used for — have no machine-readable answer attached. For rights holders that makes generation a threat; for anyone trying to license generated output, it makes it a liability.
GasLab works from the opposite premise: generation is only useful to the music business if attribution and rights data stay attached to what comes out. Not reconstructed afterwards, not asserted in a terms-of-service paragraph — attached, as data, travelling with the output.
The problem class
Making that premise real is an integration problem before it is a model problem. Rights and attribution information lives in structured sources; a generation pipeline is a fast-moving piece of AI machinery. For the two to stay connected, rights data has to be available to the pipeline as first-class context at generation time, and the provenance of an output has to be written back somewhere durable — not lost in a log file. Point-to-point glue code can fake this for a demo. It does not survive a changing pipeline, and it does not survive scrutiny from rights holders.
What we built
We worked forward-deployed with GasLab along our usual arc — audit, build, rollout, handover — and the thing we built is an integration layer on the Model Context Protocol (MCP).
At the class level: MCP servers expose the rights and attribution side to the generation pipeline as structured tools and resources, so that rights context is available where generation happens, and attribution data stays attached to the output rather than being severed from it. Because MCP is an open standard, that layer is not welded to any single model or vendor — the generation side can evolve while the rights side stays stable, which is precisely what a music-technology company needs when the model landscape shifts every quarter.
This is what MCP integration work looks like when the stakes are provenance rather than productivity: the protocol's job is to make the important data impossible to lose along the way.
Why this matters beyond GasLab
The music industry's relationship with generative AI will be decided by whether attribution survives generation. Systems that keep rights data attached make generated output something the industry can actually work with — licensable, traceable, accountable to the people whose work sits upstream. Systems that do not will stay stuck in legal limbo. We think the attached-by-construction approach wins, and building for rights holders and music companies is the core of our AI for music and rights management practice.
Ownership
As in every engagement, the layer is GasLab's: the MCP servers, the integration patterns, and the knowledge to extend them. No per-seat fees, no dependency on us. The company owns the mechanism that makes its central promise — generation with attribution — true in the data, not just in the pitch.