A standard for describing how to use L402 endpoints

[rolznz]

In https://402index.io/ there is a large list of L402 endpoints. But just the URL is not enough to teach the agent how to actually use the service.

• Description of what the endpoint does
• What HTTP method?
• What POST body
• What query params?

In x402 there is the bazaar extension, and for MPP I think there is openapi.json. Is there anything like this for L402? or should the agent go to llms.txt on the site where the endpoint lives and see if that gives the necessary info?

Any suggestions/recommendations? I think this is the biggest point of friction for agents to use L402 right now.

[Roasbeef]

See #27

[DrVelvetFog]

Can offer the x402 side of this, since bazaar got named — I run an x402 facilitator and have spent a lot of time in that spec's orbit.

What bazaar actually gives you maps closely to your bullet list: the resource server embeds a discovery blob in the 402 response itself (extensions.bazaar) with the HTTP method, input schema (query params or body), output schema, and a description — plus an MCP variant where the unit is (server URL, tool name). The structural catch is that it's still reactive: the metadata rides on the challenge, so somebody has to hit the endpoint once before anything can be cataloged, and in practice discovery happens through facilitator-run indexes built from that, not from the origin. That indirection is where most of bazaar's security text ended up (catalog poisoning, hostile metadata), which is worth keeping in mind for anyone building an L402 index like 402index.io.

#27 answers your friction point more directly than bazaar does, honestly — the OpenAPI 3.1 profile with x-l402-* extensions is exactly the "method + body + params + price in one fetch" you listed, served proactively at a well-known URI instead of piggybacked on a challenge. I left a fuller review over there (including what I'd borrow from the bazaar experience for the indexed/aggregator case). For the interim gap you're describing — endpoints that exist today with no manifest — llms.txt is a reasonable fallback but nothing enforces its accuracy; a signed manifest fetched from the origin is the only shape I've seen hold up once third-party indexes enter the picture.