We want to make agentic commerce a normal part of every developer's stack.
Agentic commerce — agents finding products, comparing them, paying for them on a user's behalf — is happening. The protocols are landing (MCP, ACP, UCP), the payment rails are starting to add agent-aware primitives, and the merchants who'll matter in this world are the ones whose catalogs an agent can discover and transact against through those protocols. We're building Steelyard so developers don't have to wire all of that up from scratch.
What we're trying to make easier
One defineCommerce manifest serves every read surface — commerce.json, a plain HTTP API, MCP, ACP, UCP — from a single call. No per-protocol boilerplate. No re-modeling your catalog three times to keep three specs happy.
UCP checkout is adapter-neutral. The merchant advertises accepted instruments; the buyer matches by instrument type. Stripe ships first, and other UCP rails can be added through the public PSP adapter contract. ACP remains the direct Stripe SPT path.
The buyer side is local-first. The wallet, the policy engine, and the vault all live on the buyer's device. Private keys never leave the keychain. We'd rather build something that works without a hosted account than something that doesn't.
And the developer experience matters. npx steelyard init scaffolds five generated Next.js route files, with the dev inspector live at /steelyard. If a dev gives up before they see a working request, the rest of the work doesn't matter.
Where it's headed
A few directions we're actively working on. These are bets, not promises — the order shifts based on what users actually run into.
- Agent-to-agent commerce.Today the buyer agent talks to a merchant. Tomorrow a buyer agent negotiates with a merchant agent that's pulling inventory from another merchant agent. The protocols are starting to support this; we want the SDK to as well.
- Multi-PSP routing.A merchant shouldn't have to pick one processor forever. We want the manifest to advertise several, and the buyer to pick by capability — cheaper fees here, faster settlement there.
- Richer mandates. AP2 is a start. We want signed deltas (cart updates that prove who edited what), recurring authorisations, and revocation flows that survive a compromised device.
- Deeper framework support. Next.js is the first framework we wired up; Remix, SvelteKit, and Astro are next. The adapter shape is small on purpose so each one is a few hundred lines, not a fork.
- Shopify and WooCommerce.The biggest e-commerce platforms run on a server stack that's not Node. A cloud-side adapter or a PHP port is the path; we'll know more as the post-v0.10 Next.js and policy work settles.
Open by default
Steelyard is MIT-licensed. The protocol adapter contract in steelyard/psp follows additive-only semver for minor releases: new optional members can appear, but breaking interface changes require a major bump. Issues, RFCs, and adapter discussion happen in the open on GitHub. If you're building in this space, we'd rather work alongside you than lock anyone in.