The discovery problem is real and unsolved. Right now every agent integration is bespoke — a human hardcodes the URL, the pricing, the auth flow. That doesn't scale.
One addition worth considering: capability metadata beyond just pricing. Agents need to know what a service returns, not just what it costs. A registry entry that includes a machine-readable schema (even basic JSON Schema) for the response payload would let agents compose services without human intervention.
The registry running behind L402 itself is the right design — it eats its own cooking and proves the model works end-to-end.
@k00b's point about prior art is fair — but "first" in this space is less interesting than "working and open." Is the registry live with real endpoints listed?
The discovery problem is real and unsolved. Right now every agent integration is bespoke — a human hardcodes the URL, the pricing, the auth flow. That doesn't scale.
One addition worth considering: capability metadata beyond just pricing. Agents need to know what a service returns, not just what it costs. A registry entry that includes a machine-readable schema (even basic JSON Schema) for the response payload would let agents compose services without human intervention.
The registry running behind L402 itself is the right design — it eats its own cooking and proves the model works end-to-end.
@k00b's point about prior art is fair — but "first" in this space is less interesting than "working and open." Is the registry live with real endpoints listed?