Introduction

Rialto is an x402 payment facilitator and Bazaar discovery layer for Stellar - so AI agents can find, pay for, and verify paid services on stellar:testnet and stellar:pubnet.

The problem, measured

x402 lets software pay for HTTP resources with an on-chain payment instead of an API key. On Stellar, settlement already works - we verified it ourselves with real settled transactions through the reference tooling and the official end-to-end test suite. What is missing is discovery: an agent can only pay endpoints it was already told about.

We pulled the main public x402 catalog (Coinbase's Bazaar - 14,107 resources) in full and measured Stellar's actual position in it:

• 518 entries advertise Stellar - but 92% belong to a single seller listing Stellar alongside every other chain
• the remaining entries use malformed asset identifiers no conformant Stellar client can settle
• zero entries are Stellar-only, zero offer metered (upto) payments on Stellar, and zero MCP tools are cataloged in it
• the catalog's own search ignores network intent: querying for a Stellar service returns EVM results

Stellar is technically listed and practically invisible. Rialto exists to fix that.

What Rialto builds

Built and tested end to end - not a plan, a running system:

1. A facilitator - canonical /verify, /settle, /supported on stellar:testnet, composing the Apache-2.0 @x402/stellar package rather than reimplementing settlement. exact scheme only so far; upto (below) is spec-only.
2. A discovery catalog - services are indexed automatically the moment a payment settles; no registration step. Validated against poisoning (the integrity gauntlet), searchable with real hybrid ranking - keyword (BM25F) fused with semantic search (embeddings) - and ranked by real settlement history. Backed by a public, re-runnable eval harness with real numbers, not an unverified claim.
3. Federation - independent facilitators can register, external catalogs are ingested and cross-published: a service settles anywhere but is findable everywhere.

Still design-stage or spec-only, stated plainly:

4. The upto payment scheme for Stellar - the full spec is written; the Soroban contract and facilitator integration aren't built yet.
5. An MCP server - so agents will eventually run discover → pay → verify from inside their own runtime. Today, the discovery HTTP API is callable directly (see the buyer/agent guide) while the MCP wrapper is built.

Where to go next

• Architecture - components, flows, trust boundaries, diagram
• How search works - the actual ranking math, derived
• Threat model - trust boundaries, and one real bug found and fixed during its own research
• upto scheme spec - the full Stellar scheme specification
• Design decisions - why the cap lives on-chain, and how search quality is proven
• Guides for sellers, buyers/agents, and operators