Agentic Licensing for the Large Language Model Commons: MCP and NFT-Linked Rights Objects

Abstract

Large language models have intensified a growing property-rights challenge in digital markets: protected works can be copied, retrieved, transformed, and recombined at low marginal cost, while ownership, licensing authority, attribution, and remuneration remain costly to verify. First, I introduce the Model Context Protocol (MCP) as an interoperability layer between AI agents and intellectual-property institutions. MCP does not define rights or settle disputes; it gives agents a standardized way to query registries, invoke licensing tools, execute payments, record usage, and preserve audit trails. Second, I develop a stylized transaction-cost model of agentic licensing and derive comparative statics for when lawful exchange expands. Lower search, verification, contracting, payment, and monitoring costs can move marginal uses from avoidance, substitution, or unauthorized use into licensed exchange, especially when rights records are reliable, license terms are standardized, and interface costs are large relative to the price of the license. Third, I explain how non-fungible tokens (NFTs) can complement MCP when they operate not as collectibles, but as machine-readable rights objects linked to work identifiers, ownership claims, license scope, payment rules, provenance records, audit obligations, and dispute forums. Music licensing is illustrative because rights are fragmented across compositions, recordings, labels, publishers, performers, territories, and use types. MCP and NFT-linked rights records can support ex ante licensing when paired with verified title, enforceable contracts, bounded delegation, human review, and off-chain legal remedies.