Callisto is a gateway that lets an AI assistant buy from your BigCommerce store the way a company buyer does, inside the rules you already run.
The account's own price list, its credit and terms, a purchase order, and the approval chain you already run in B2B Edition.
Think about how your repeat business actually arrives. A purchase order attached to an email. A phone call to the order desk. The same buyer signing into the portal to rebuild the same forty lines they ordered last month, while someone on your side re-keys it and the error rate is whatever it is.
That work is repetitive and rule-bound, which is exactly the kind of work a buyer hands to an assistant first. Some procurement teams are already building bots to do it.
There is already a standard for this. Google launched the Universal Commerce Protocol in January 2026, and the platforms and payment networks signed on. But UCP was designed around consumer checkout: a card, one shopper, a list price. Your accounts don't work that way. They have a negotiated price list, a credit limit, net terms, a PO number, and a junior buyer who isn't allowed to spend past a threshold without the purchasing manager seeing it first.
So an agent that reaches your store today either can't finish the order, or finishes it at the wrong price with none of those controls attached.
When a buyer's assistant can reorder from one supplier in a sentence and has to open a portal for the other, the first supplier becomes the default.
A junior buyer asks for 170 hoodies. Contract pricing, a guardrail, a quote negotiated with a sales rep and an approval all happen before the order reaches BigCommerce.
How one order moves through Callisto, from the buyer's sentence to the order record.
Your storefront doesn't change and your catalog, customers and orders stay where they are. The gateway reads from BigCommerce, applies the account's rules, and writes the finished order back.
The assistant signs the buyer in. Claude connects over MCP, and any agent that speaks UCP connects over REST.
Every request is made as one named company user. Price, credit, approval and guardrails are applied before anything is placed.
The order arrives in your store looking like any other order, with a staff note naming the agent.
Setting it up against your store is scripted. One command checks what it needs, read-only, and reports back before anything is written.
There is a version of this that is easy to build and dangerous to run. Hand an agent an admin key, point it at the store, and it can order anything, at any price, for any customer, with nothing in the way.
This works the other way round. A token belongs to one company user and carries that user's role. The price comes from the account's price list rather than from the conversation. Credit is checked when the order is made and again when it is placed, because an approval can sit for days. An order that trips a guardrail goes to a person, and never back to the person who asked for it.
Trust comes from the rails, not from the agent being clever.
UCP is not our protocol. Google launched it in January 2026 as an open standard for agentic commerce, co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed by more than twenty others including BigCommerce's parent company, Visa, Mastercard, Stripe and Adyen. The platforms and the payment networks have agreed on how an agent should be able to buy.
What they agreed on was consumer checkout. UCP carries no company accounts, no contract pricing, no terms, no purchase orders, no approvals, quotes or reordering. We wrote those as seven extensions under our own namespace, implemented them against a live store, and published the specs for anyone to build against. Where UCP has an open proposal covering the same ground, we follow its field names so the work moves across if that proposal is adopted.
Which means you are adopting an agreed standard, plus a published extension to it. You aren't buying a private API that only we can speak.
The gateway runs today on Wildleaf, our own B2B Edition demo store, stocked with fictional companies and buyers. It has placed real orders end to end: contract pricing applied, a guardrail cleared, a quote negotiated with a sales rep, an approval granted by the account's admin, and the order sitting in BigCommerce on purchase order terms.
The extension specs are drafts. They are implemented and tested against UCP's published schemas, and they will keep moving as UCP does.
We want a merchant to run the first pilot with us. On your side that means a few accounts willing to try it, somebody who can answer questions about how your pricing and approvals really work, and a store we can connect to.
Commerce Studio is a senior commerce practice built on twenty-five years of commerce platform programs and led by the author of Agentic Commerce (2026). Callisto is the working proof of that book's argument: the next buying interface is an agent, and B2B needs rules before it needs intelligence.
Less of a fit: sellers whose B2B ordering already runs on EDI or punchout and works. This is for the accounts that still order by portal, email and phone, which in most distributors is the longer half of the list.
MCP is one way an agent connects, and we offer REST as well. Behind that connection sits a gateway that knows each buyer's contract price, credit and approval chain, and enforces all three. That is the part that matters.
Good. We want them to, and we are building on the same standard they endorsed. The native work is consumer checkout: cards, one shopper, list prices. Business buying needs accounts, contract prices, terms, purchase orders and approval, which is what our extensions add on top. As the platform adds B2B pieces of its own, Callisto layers onto them.
The agent acts as one user with that user's role. Credit is checked twice. A junior buyer's order goes for approval. Company guardrails cap each agent order and the month's agent spend, and flag quantities well outside the usual. Anything that trips goes to a different person, and all of it is in the audit trail.
Some of them are, and procurement teams are building their own bots. A pilot with two or three friendly accounts costs very little, and it puts you on the list of suppliers that are easy to buy from when the rest of them start.
No. Everything stays in BigCommerce and arrives the way it always has. Nothing changes on the theme.
The gateway needs API access to read prices and place orders, the same as any integration, and it runs as a deployment dedicated to your store. Buyers' agents never see those credentials. They get narrow, short-lived tokens scoped to one user.
A handful of friendly accounts and a real agent placing real orders against your catalog. If that is useful, we can work out what a pilot looks like on your store.
Talk to us about a pilot