What the Universal Commerce Protocol actually is

UCP is an open standard for agentic commerce. It gives AI agents a common language for discovering a merchant’s products, building a cart, applying the merchant’s own pricing and promotions, and completing a checkout, without the merchant having to build a bespoke integration for every assistant that comes knocking.

Think of it as the difference between a product feed and a storefront API.

A feed is a static export: here are my SKUs, my prices, my images. Whatever the receiving platform does with that data is out of your hands. It goes stale. It does not know about your bundle logic, your tiered discounts, your inventory by location, or the fact that this particular SKU is available on preorder only.

UCP is designed to be the opposite. It is a live surface that the merchant controls. The agent does not scrape a cached copy of your catalogue. It queries your commerce system in real time and gets back what your storefront would have returned. Prices are current. Inventory is current. Your promotion engine still runs. Your fulfilment rules still apply.

The strategic claim underneath it is straightforward. Agentic commerce is happening whether merchants like it or not, so the standard that governs it should be one where the merchant keeps the customer relationship, the brand presentation, the margin control and the transaction data, rather than handing all four to whichever assistant intermediated the sale.

Why a protocol was needed at all

Without a standard, agentic commerce degrades into one of three bad outcomes for brands.

One: scraping. Agents parse your storefront HTML like a very fast, very literal shopper. They get prices wrong, miss variant logic, ignore regional pricing, and load your servers. You have no visibility and no control.

Two: marketplace capture. Every assistant builds its own closed commerce layer. To be visible inside it you list your products on their terms, at their take rate, with their checkout, and their customer record. This is the Amazon dynamic, rebuilt for the AI era, and it is the single largest structural risk to D2C economics in a decade.

Three: integration sprawl. Every assistant ships a different API and you build to each one separately. Five surfaces, five integrations, five sets of edge cases, five things that break during peak season. Only the largest merchants can afford it, so everyone else is invisible.

A shared protocol collapses all three. Build once, be reachable everywhere, keep control of what happens.

The moving parts

Implementations vary, but the architecture of agentic commerce standards generally resolves into four layers. It is worth understanding them separately, because they fail separately.

1. Discovery

The agent needs to know what you sell and whether it is relevant. This means structured, queryable access to your catalogue. Not just titles and prices, but the attributes that let an agent reason: materials, dimensions, compatibility, care, country of origin, lead time. The richer and cleaner your product data, the more often you become the correct answer.

2. Context and negotiation

The agent needs to resolve a real, purchasable configuration. Which variant is in stock? Does this discount code stack? What is shipping to this pincode, and by when? This is where a live protocol beats a feed decisively, because the merchant’s own logic answers the question.

3. Cart and checkout

The agent constructs an order against the merchant’s commerce system. In a model where the merchant retains control, the order is created in your system, with your order number, your fulfilment workflow, and your customer record, even though the interaction happened on someone else’s surface.

4. Payment and authorisation

The hardest layer, and the one attracting the most separate standards work. It has to answer: did the human actually authorise this, for this amount, for this item, within these limits? Who is liable if the agent gets it wrong? How does a chargeback work when a machine pressed the button?

UCP, ACP, AP2 and MCP: an honest map

There is a genuine standards contest happening here, and the acronyms get muddled. A rough map:

MCP (Model Context Protocol) is the substrate, not a commerce standard. It is how AI models connect to external tools and data sources generally. Several commerce protocols are built to sit on top of it. If you have heard someone say they exposed their catalogue over MCP, this is the layer they mean.

ACP (Agentic Commerce Protocol) came out of the OpenAI and Stripe camp and underpins purchasing inside ChatGPT. It solved a real problem early and got real transaction volume, which matters.

AP2 (Agent Payments Protocol) is Google originated and focused specifically on payments and authorisation, proving that a human granted a mandate to an agent to spend money under defined constraints. It complements protocols that cover catalogue and cart rather than competing with them.

UCP positions itself as the standard that sits on the merchant’s side: the one that treats the brand’s commerce system as the source of truth and the assistant as a distribution surface, not an owner.

The likely outcome is not a single winner. It is a stack, with a context layer, a commerce layer and a payments layer, more than one option at each level, and merchants supporting whichever combinations their customers actually use. Betting the business on one protocol right now would be premature. Ignoring all of them would be worse.

What genuinely changes for merchants

Product data becomes a ranking factor, not a housekeeping task. An agent choosing between three jackets will pick the one whose attributes let it answer the shopper’s actual question. Thin descriptions, missing specs, inconsistent variant naming and image only spec sheets make you unpickable. This is the least glamorous and highest leverage work available right now.

Merchandising loses some of its levers. No hero banner, no carefully sequenced collection page, no exit intent offer. Persuasion moves upstream into the data itself: reviews, structured claims, comparison friendly attributes, clear returns and shipping terms.

Attribution gets harder before it gets easier. Sessions that begin inside an assistant carry less of the signal your analytics stack expects. Expect a growing bucket of demand that looks like direct traffic but is not, and plan measurement accordingly.

What happens after the purchase becomes the brand. If the discovery and checkout moment is increasingly rented, the parts you still fully own carry more weight than they used to. Packaging, unboxing, the order confirmation, the WhatsApp update, the returns experience, the second purchase.

Returns and disputes need a policy update. “My agent bought the wrong size” is a support ticket category you do not have a macro for yet.

The India layer nobody is discussing

Most commentary on agentic commerce is written for a prepaid market with cards on file and instant checkout. Indian D2C is not that.

Three complications worth naming:

Cash on delivery. A meaningful share of Indian D2C orders still settle at the door. An order placed by an agent and paid on delivery has an entirely different risk profile. No payment friction means no self selection, and RTO rates on unverified agent orders could be brutal. Any brand enabling agentic checkout should be thinking about how to gate COD from day one.

Checkout layers from third parties. GoKwik, Shiprocket Checkout and similar tools sit between the Shopify cart and the payment, and they carry a lot of the conversion logic: address intelligence, RTO scoring, UPI flows, prepaid nudges. Agentic checkout protocols assume a relatively clean path from cart to payment. Where that path runs through another vendor’s layer, the integration story is unresolved and worth asking about directly.

UPI and mandates. Payments authorised by an agent map more naturally onto mandate based rails than onto cards on file. India already has sophisticated mandate infrastructure. It is plausible that agentic payments here evolve along a different track than the US model, and that brands built on UPI first checkout end up better positioned, not worse.

The risks, stated plainly

We would be doing you a disservice by presenting this as unambiguous upside.

Disintermediation is still the base case. A protocol that keeps the order in your system is meaningfully better than a marketplace listing. It is not the same as owning the demand. The assistant still decides which three products to surface, and that ranking logic is not yours.

Standards churn is real. Some of what is being announced this year will be deprecated within eighteen months. Building deep, custom infrastructure against a first release is a way to buy technical debt.

Fraud and abuse vectors are immature. Automated purchasing at machine speed against limited inventory is a nightmare for any brand that runs drops, and the bot detection most stores rely on was not designed to tell a legitimate shopping agent apart from a scalper’s script.

Discovery may consolidate rather than democratise. The optimistic reading is that great product data lets a small brand compete with a large one. The pessimistic reading is that assistants default to recognisable brands and high review volume, and agentic commerce concentrates demand further. Both are plausible. We do not yet know which is true.

What to actually do this quarter

A pragmatic sequence, roughly in order of return on effort:

1. Audit your product data like it is a technical asset. Complete, consistent, structured attributes across the catalogue. Variant naming a machine can parse. Specs in text, not baked into images. This pays off in classic SEO, in paid media feeds, and in agentic discovery at the same time. There is no scenario where it is wasted work.
2. Fix your structured markup. Product, Offer, AggregateRating, shipping and returns schema, implemented properly and validated. Boring, unglamorous, still the cheapest visibility work available.
3. Publish the answers agents need. Sizing guidance, materials and care, comparison content, clear returns and delivery terms, written to be quoted rather than to fill a word count.
4. Talk to your platform and checkout vendors. Ask Shopify, and ask your checkout provider specifically, what their agentic commerce roadmap is and what it will require of you. Get the answer in writing.
5. Decide your COD and fraud posture before you need it. If orders placed by agents start arriving, what is your rule?
6. Instrument what you can. Start segmenting traffic referred by assistants now, even crudely, so you have a baseline when the volume becomes material.
7. Do not rebuild your stack yet. Readiness is data hygiene and vendor alignment. It is not a large integration project against a spec that will change.

Our view

The Universal Commerce Protocol matters less as a piece of technology than as a statement of position: that in agentic commerce, the merchant’s system should remain the source of truth.

Whether UCP specifically wins is genuinely uncertain. What is not uncertain is the direction. Discovery is moving into conversational surfaces, and the brands that arrive with clean, complete product data that machines can read will be the ones those surfaces can actually recommend. The brands with thin descriptions and specs trapped inside JPEGs will simply be invisible. Not penalised, just unreadable.

That work is available to do today, costs nothing but discipline, and pays off regardless of which acronym wins.

Frequently Asked Questions: UCP

What is the Universal Commerce Protocol in simple terms?

An open standard that lets AI assistants browse a merchant’s real catalogue, build a cart and complete a purchase using the merchant’s own pricing, inventory and checkout rules, rather than a stale product feed or a scraped copy of the storefront.

Is UCP the same as ACP?

No. They are separate agentic commerce standards from different camps, addressing overlapping problems. Most brands will end up supporting more than one.

Do I need to do anything technical right now?

Probably not a build. The highest value preparation is product data quality, structured markup, and a conversation with your platform and checkout vendors about their roadmap.

Will this replace my website?

No. It adds a surface. Your storefront remains where brand, considered purchases, repeat customers and higher value baskets live.

How does this affect paid media?

Feed quality already determines Shopping and Performance Max outcomes. The same catalogue discipline that improves agentic discoverability improves feed performance. One investment, two returns.

NOIR & BLANCO builds and scales Shopify storefronts and paid media programmes for premium D2C brands. If you want a read on how ready your catalogue actually is for agents, [get in touch].