Data marketplace for agents
MCP-native discovery and per-call metering for data products.
An autonomous agent needs data to act — a pricing feed, a geospatial layer, a model output, an API, a slice of a training corpus. Today it gets that data through bilateral integrations and annual contracts negotiated by humans. That doesn't scale to thousands of agents making thousands of decisions an hour, each needing a different source for a few seconds.
Fabric makes data a tradable primitive an agent can handle end to end in a single, signed network conversation: discover the right product, verify its quality and licence, pay per call, and consume it — no human in the loop, no standing contract. Discovery is exposed as MCP tools, so the agent treats a data market the way it treats any other tool.
What changes
Acquisition shifts from annual contracts to spot, per-call markets — an agent buys exactly the data it needs, when it needs it.
Lineage is signed — consumers can prove what source they used, at what version, for which decision.
New data providers reach autonomous customers without an enterprise sales motion — publishing a product is the go-to-market.
Agents compose sources at runtime based on observed quality, freshness, and price, switching providers without re-integration.
Licensing and purpose are enforced as credentials — usage outside the agreed purpose is verifiable and refusable, not a contractual afterthought.
Where to start
Wrap one existing data API as a Fabric-published offering, add a credential for quality and freshness, and wire micropayment-per-call via Tokenisation. Expose discovery as an MCP tool and have a test agent find, license, and call it end to end. The hardest design call is usually the licence-and-purpose credential model — solve that first.