x402 & the Agent Economy (Concept)

Concept in progress. We're still working out how payments, MCP delivery, and discovery fit together — none of this is wired up yet. Pricing specifics are captured separately in Pricing & Cost Model.

Why this matters

When the primary API consumer is an autonomous AI agent, not a human, the whole commercial model changes. Agents don't sign up, manage API keys, or type in card numbers — but they can carry wallets and pay programmatically. The rails for that have matured recently, which is why we're exploring them now rather than inventing bespoke billing.

x402 — payment (planned approach)

x402 revives the dormant HTTP 402 "Payment Required" status code as a standard for machine-to-machine payments.

The flow we're designing toward:

1. Agent requests a simulation.
2. Server replies 402 Payment Required with a price (e.g. USDC on Base), quoted dynamically based on the request.
3. Agent signs a gasless stablecoin transfer authorization and retries with the signature in the header.
4. A facilitator verifies and settles the payment on-chain; the server then runs the simulation and returns the result.

Why it fits: permissionless, account-less, and built for agents paying for API access. We'd lean on existing middleware and facilitators (e.g. Coinbase/Stripe) rather than building settlement ourselves — but we haven't integrated any of this yet.

MCP — integration (planned delivery shape)

A bare REST API alone isn't enough: an LLM can't reliably guess how to call complex endpoints. Our current thinking is to deliver through an MCP server (see Agent API & MCP (Concept)) so agents get explicit, self-describing tools — the "agent bindings" that make the API usable. Still at the design/sketch stage.

Discovery — registries (future GTM)

Agents would find tools through MCP registries (e.g. the official registry.modelcontextprotocol.io, plus gateways/marketplaces), and enterprises would mirror tools into private registries. Discovery is real but not magic: it rewards precise metadata and good docs, and humans still allowlist tools for high-stakes tasks — so registry presence would pair with active distribution once we have something to list.

Timing — why explore this now

We're in a "Goldilocks window" in mid-2026: the standards have hardened (MCP for how agents talk, x402 for how agents pay, ERC-8004 for agent identity), facilitators are in production, but the high-value, domain-specific endpoints — like a deterministic EVM oracle — mostly don't exist yet. Build a year ago and you'd be writing the plumbing; wait a year and the registries fill up. That window is why we're conceptualizing now even though the product isn't built.

A note on identity & anti-abuse

ERC-8004 (trustless agent identities) and simple on-chain checks (e.g. a minimum wallet balance) could give cheap, effective spam resistance without human-style signups — relevant to how we'd keep the service open yet protected (see Pricing & Cost Model). Directional only; not implemented.

Changelog

x402 Concept Changelog

1. v0.12026-06-30Initial outline — x402 payment flow, MCP as delivery layer, registry discovery; reframed as planned, not shipped.

Add a one-line entry here whenever this concept changes.