Magento agentic commerce: extension or service?

There is a real question behind this page, and it deserves a real answer rather than a sales one: an extension that installs a UCP manifest, an MCP endpoint and an llms.txt costs between $99 and $399 and takes an afternoon. A readiness sprint costs from $4,500 and takes five business days. If both end with the same artifacts on the storefront, only one of them makes sense.

They do not end with the same artifacts, but the difference is not the one vendors usually claim. Here it is stated plainly.

What an extension actually does, and does well

The Magento extension market covered agentic commerce fast. As of August 2026 you can buy, off the shelf:

• Protocol endpoints — Mirasvit's Agentic Commerce module ($169 first year, $102/year after) plus its MCP integration ($149): UCP manifest, MCP server, WebMCP, ACP feed, llms.txt, agents.md, agent JSON APIs. Meetanshi ships a comparable set including ACP checkout sessions with HMAC.
• Feeds — Plumrocket's ChatGPT product feed at $99 lifetime, Magmodules' OpenAI commerce feed in the same band.
• llms.txt generators — a cluster of modules at $49–$149.
• Structured data — WeltPixel and Amasty rich-snippet modules, $0–$99.
• Free and open source — Magebit publishes ACP and UCP modules at no cost, and we publish magento-agent-readiness under MIT: in-admin readiness diagnostics, no licence, no upsell in the code.

This is not a bad market. The protocol code is converging on free, and that is the correct destination for it — writing a UCP manifest is not the hard part of agentic commerce and never was. If your store has a clean catalog, complete product data, a sane robots policy and one developer who reads specifications, an extension plus a day of that developer's time is a perfectly good outcome. Buy the extension. You are done.

Where the extension stops

An extension installs endpoints. It cannot make the data behind them true, and it does not answer for the result. Four things sit outside every module we have tested:

1. The catalog underneath. In our Q3 2026 benchmark of 1,854 live Adobe Commerce and Magento storefronts, 54% of the product pages an agent can reach carry no usable Product + Offer structured data. A schema module writes the block; it does not fill in the GTIN that was never in the PIM, reconcile the price that ERP and storefront disagree about, or decide what a configurable product looks like to an agent that wants one buyable SKU. Bad data rendered in perfect JSON-LD is still bad data — and the agent, unlike the shopper, has no way to squint past it.

2. Programme eligibility. OpenAI's merchant programme is US-only. Google's Merchant Center and Perplexity's merchant programme each have their own onboarding, their own feed requirements and their own rejection reasons. No Magento module gets you through them; that work is applications, feed corrections and waiting, and someone has to own it. See ChatGPT Shopping for Magento: US and EU.

3. The edge. Most stores with a bot problem solved it years ago with a WAF rule, and those rules do not know the difference between a training crawler and a shopping agent about to send a customer. Cloudflare's 2026 crawl-to-refer ratios explain why merchants got defensive — ClaudeBot at roughly 50,000 crawls per referral, GPTBot at 887:1, Perplexity at 118:1, against about 4.9:1 for Google — and Akamai measured in July 2026 that 47.9% of AI bot traffic targets commerce. Installing an MCP endpoint behind a WAF that returns 403 to the agents you built it for is the single most common failure we find. No extension can see this, because from inside Magento the endpoint looks fine.

4. Standards churn. OpenAI retired Instant Checkout in March 2026, about five months after launching it. UCP consolidated the other direction — announced at NRF in January 2026, joined by Amazon, Meta, Microsoft, Salesforce and Stripe in April. A module bought today reflects the map as of its release. Somebody has to watch the map.

The honest split

Your situation	What to buy
Clean catalog, in-house dev, you want the endpoints to exist	An extension, or the free module. Not a sprint.
You want to know where you actually stand, first	The free scan — eight checks, no charge, no call
Endpoints plus feeds, share-of-voice measurement and schema repair, maintained	Agentic Commerce Suite Pro — $399 once per installation
Messy catalog, ERP/PIM in the loop, B2B pricing, custom checkout, or nobody to own the outcome	The Readiness Sprint — from $4,500
You need somebody accountable to a board for "are we visible in AI shopping"	The sprint, then monitoring

We sell all three tiers, which is exactly why this table is worth reading: the row that sends you to a competitor's $169 module is in it because sending the wrong buyer into a $4,500 engagement costs us more than the sale is worth.

What the sprint is actually for

The sentence we would defend in front of a technical buyer: an extension installs endpoints; the sprint validates whether agents can actually read your store and transact with it. Five business days, fixed scope, one price published on the page: an audit of catalog and commerce data against what agents need, an architecture aligned to ACP, UCP and MCP with no replatforming (Magento, ERP, PIM and payments stay the source of truth), a working prototype of one critical path on your real stack, security and observability, and a production roadmap with estimates. If the answer at the end is "install these three modules and fix these eleven attributes," that is the deliverable and the roadmap says so.

The proof is not a case study slide. Nord Trail is our own reference storefront: every artifact the scanner checks exists there for real, and it is re-scanned every morning by the same public scanner we point at your store. Run the scan on it yourself.

Questions

Is $4,500 not simply 25 times too expensive for the same result?

For the same result, yes — and if that is your situation, buy the module. The price buys judgment and accountability on a store where the result is not the same: which pieces, in which order, against which data problems, and a named engineer who signs the architecture. For a mid-market merchant paying an agency $2,000–$8,000 a month in maintenance, the sprint is under a month of retainer.

Can we start with the extension and hire you later?

Yes, and it is the sequence we recommend for a store with an in-house developer. Install the free module or a paid one, run the free scan, and see how far that gets you. If the scan comes back clean, you did not need us. If it comes back with catalog or edge problems a module cannot reach, you now know exactly what you would be buying.

Do you resell or take a cut of the extensions you recommend?

No. We recommend third-party modules by name where they are the best fit, including modules from vendors we compete with, and we take nothing for it. Our own module is priced on its own page and the free tier is MIT-licensed on GitHub.

Will you guarantee we appear in ChatGPT or Google AI Mode?

No, and nobody who does should be believed. Discovery, ranking and programme access are controlled by OpenAI, Google, Perplexity and Adobe. We can guarantee that your store is readable, eligible where eligibility exists, and not blocked at the edge — that is the whole of what any vendor controls.