Adobe Commerce MCP: what works today, and when waiting is the right call

"Adobe is building this — we will wait for the platform" is a defensible position, and on several points it is the correct one. It is also, for most Magento stores, an answer to a question that was not asked: much of what makes a store readable to shopping agents sits outside anything Adobe has announced, and it is not scheduled to arrive in the platform at all.

This page separates the three.

What Adobe has announced

• 18 February 2026 — Adobe stated support for the Universal Commerce Protocol (UCP) and the Agentic Commerce Protocol (ACP).
• Adobe Summit, 20 April 2026 — Adobe announced the Commerce MCP Server (catalog, cart, pricing, inventory, promotions, checkout, orders exposed as MCP tools), a Commerce Developer Agent, and Brand Concierge.

That is a serious roadmap and it points the right way. What it is not, as of this page's review date, is something you can switch on.

What is generally available today

The Commerce MCP Server is tied to App Builder and the Integration Starter Kit and is described as arriving "as soon as generally available," with no published GA date. The fuller agentic capabilities lean on Adobe Experience Platform, which is a separate licence. The centre of gravity is Commerce as a Cloud Service — Adobe's SaaS edition.

Three consequences follow, and they decide whether waiting is rational for you:

1. Magento Open Source and self-hosted Adobe Commerce do not get this automatically. The largest part of the installed base is not on the road Adobe is paving. If you are self-hosted, waiting is waiting for something that is not coming to your address.
2. If you are on Cloud Service or heading there, waiting for the MCP server is reasonable — for the MCP server specifically. Do not rebuild what the platform will hand you for free in a few quarters.
3. AEP-gated capability is a budget question, not a technical one. The objection we hear most often — "on Adobe Commerce this is a six-figure project" — is true for the AEP path and false for everything below it. Knowing which one you are being sold matters more than any protocol detail.

What Adobe's roadmap does not cover

An MCP server exposes your catalog and cart to an agent that has already decided to talk to your store. Everything that determines whether an agent gets that far, or finds anything worth reading when it does, is yours regardless of edition:

Product data quality. In our benchmark of 1,854 live Adobe Commerce and Magento storefronts, 54% of reachable product pages fail the structured-data check and 48% carry no Product JSON-LD at all. A platform MCP server serves whatever is in your catalog. Missing GTINs, ERP-versus-storefront price disagreements and configurables that do not resolve to a buyable SKU stay missing, disagreeing and unresolved.

Discovery feeds. Google Merchant Center for AI Mode and Gemini, the Perplexity merchant programme, the OpenAI product feed — these are how agents find you in the first place, they are live now, and they are not part of Adobe's MCP story. Discovery works today; agentic checkout is the part still in flux.

Bot policy at the edge. Your WAF, your CDN and your robots.txt decide whether an AI shopping agent is served or shown a 403, and they were configured before shopping agents existed. Cloudflare's 2026 crawl-to-refer figures (ClaudeBot around 50,000:1, GPTBot 887:1, Perplexity 118:1, Google about 4.9:1) explain why merchants clamped down; Akamai measured 47.9% of AI bot traffic pointed at commerce in July 2026. Adobe does not manage your edge, and a blocked agent never reaches the MCP server at all.

Measurement. Roughly 71% of ChatGPT-referred visits land in GA4 as Direct, and Google's AI Mode deliberately sends noreferrer. GA4's "AI Assistant" channel, added on 13 May 2026, classifies on the Referer header and therefore understates by construction. If you cannot see the channel, you cannot defend the budget for it — the answer is server logs as the source of truth plus differential measurement on SKU cohorts, and no platform release will do this for you.

The cost of waiting, stated fairly

The real risk is not that Adobe ships late. It is that "we are waiting for Adobe" quietly covers four unrelated pieces of work that were never Adobe's, and they compound: catalog debt takes months of PIM work, programme eligibility takes weeks of calendar time you cannot compress, an edge rule takes a day but nobody owns it, and measurement is retroactively impossible — data you did not collect in 2026 cannot be collected in 2027.

Set against that, keep the volume honest. Peer-reviewed measurement (Kaiser & Schulze, Marketing Science 45(4), April 2026; 973 stores, 6 billion sessions) puts LLM-referred traffic below 0.2% of total traffic, converting above paid social and below organic search. Checkout.com reported in June 2026 that around 3% of transactions involve an agent. This is not a channel that will save a bad quarter, and anyone telling you otherwise is quoting a growth rate without a denominator. It is a channel worth being readable in, cheaply, before it matters.

What we would actually do in your position

Run the free scan — eight checks, no charge, no call — and read the result against the list above. If everything it flags is protocol plumbing and you are on Cloud Service, wait for Adobe and spend the quarter on your catalog instead. If it flags edge blocking, structured data or feeds, none of that is on Adobe's roadmap and waiting costs you the whole delay.

If you would rather have someone own it: the Readiness Sprint is five business days at a published price, aligned to ACP, UCP and MCP without replatforming, and it is explicitly designed to survive the platform shipping — the deliverable is your data, your eligibility and your architecture, none of which Adobe's GA date changes. Our own reference storefront, Nord Trail, passes all eight checks on plain Magento and is re-scanned every morning.

Questions

Will the Commerce MCP Server make the sprint redundant?

For the endpoint layer on Cloud Service, largely yes, and we would rather say so now than sell around it. What it does not touch is catalog quality, merchant-programme eligibility, edge policy and measurement — which is most of the work and all of the durable part.

We are on Magento Open Source. Is anything coming for us?

Nothing announced. The protocol code is already free from several sources, including our own MIT-licensed module; what is not free is the data and eligibility work behind it.

Should we replatform to Cloud Service or Shopify to get this natively?

Not for this reason alone. Shopify does bundle agent readiness into the platform, and that is a genuine advantage — but a replatform costs more than a decade of the readiness work it saves, and your catalog debt migrates with you.

Is UCP or ACP going to win?

UCP has consolidated faster: announced at NRF in January 2026, with Amazon, Meta, Microsoft, Salesforce and Stripe joining through April. OpenAI retired Instant Checkout in March 2026, roughly five months after launch, which moved ACP toward discovery. We build to both and we do not need the question answered — the catalog readability underneath is the same either way, which is precisely why we sell that and not a bet on a protocol.