Pixee Commerce: An E-Commerce Platform Built for the Agentic Era
Why we built a platform that ships MCP and the agentic protocols by default rather than as add-ons — and what that changes in practice.
The three previous articles in this series made one observation: agents are already making purchases, the protocols have stabilized, and upgrading an existing store involves a list of fixes. This article explains the conclusion we’ve drawn as publishers: on a new platform, these capabilities shouldn’t be afterthoughts.
The Problem with After-the-Fact Extensions
Most platforms today treat agentic commerce as an add-on module. This works to a certain extent, but it leads to three systematic flaws:
- Duplication of business logic. Rules for pricing, inventory, and eligibility exist once in the core system, a second time in the agentic extension, and a third time in the chatbot. They diverge within a few months.
- Latency. An extension queries the core system via a general-purpose API, often with a cache and often with a delay. This is exactly what you don’t want when it comes to inventory.
- The security blind spot. An extension that exposes write capabilities reimplements access controls in its own way. This is the most common source of security incidents.
What We’ve Built
Pixee Commerce is built on Medusa v2, an open and modular e-commerce core, which we’ve extended with the building blocks that the agentic era makes indispensable. The guiding principle: a single business logic, multiple access points.
Agentic protocols at the core
A front-end MCP server and an administration MCP server expose the platform’s capabilities—product search, availability, shopping cart, order—using the same business rules as the web interface. The ACP, UCP, and AP2 transport protocols build on this foundation. There is no need to maintain a second implementation in parallel.
Hybrid search and discovery
MeiliSearch for keyword search, Qdrant for semantic search. This is what enables the system to respond to a query phrased in natural language—such as “something lightweight for trail running, with good cushioning”—rather than a series of filters. An agent formulates its queries exactly this way.
Conversational channels as sales channels
Native Telegram and WhatsApp bots, connected to the same core. Zero duplication of logic: the shopping cart built within a conversation is the same object as the one on the website.
AI powering product pages
Storefront chat, image generation, translation. When the customer has both components, these are powered by the Pixee PIM repository, which avoids the classic scenario of a catalog that’s enriched on one side and depleted on the other.
The Three Key Decisions
You own the code
The foundation is open-source; the platform can be deployed on your own servers or with the hosting provider of your choice. This is a deliberate choice: at a time when sales channels are reshaping every eighteen months, relying on a “black box” over which you have no control—neither the roadmap nor the terms of access—is a business risk, not just a technical one.
No commission on revenue
The business model is not tied to your revenue. A platform whose cost increases in proportion to your success will always end up working against you.
Hosting in France, compliance by design
French infrastructure, personal data processing built in from the start, and the ability to route AI processing to your own providers or to a model hosted in Europe. For many of our clients, this is a contractual requirement, not a preference.
Six preconfigured verticals
The platform comes with ready-to-use industry configurations: high-tech, fashion and direct-to-consumer, beauty and skincare, jewelry and accessories, outdoor and sports, and premium food. Each provides its own attribute model, discovery components—comparison tools, product questionnaires, configurators—and editorial tone.
The benefit isn’t just cosmetic: these configurations incorporate the category’s data model. It is this model that determines whether an agent will find the information they need to recommend your products.
Who it’s for—and who it isn’t
| Situation | Our recommendation |
|---|---|
| Platform redesign planned within the next twelve months | The best time: the cost of integrating the protocols is minimal during the design phase |
| Existing store that’s running smoothly | Start with the technical checklist, not with a migration |
| Complex, multi-channel catalog | Address the product database first; the platform comes next |
| Need for a conversational sales channel | This is the use case where the gap with an added extension is most noticeable |
| Very small, single-channel catalog | A standard hosted solution remains more cost-effective |
Frequently Asked Questions
Do we need to migrate to have a presence in agentic commerce?
No, and we tell customers who call us about this. An existing, well-maintained store can be made accessible to agents in just a few weeks. Migration is justified for other reasons—functional limitations, commission costs, the need for control—and the agentic marketplace is simply an additional factor.
How long does it take to go live?
For a preconfigured vertical with a clean catalog, it takes a few weeks to go live for the first time. The key factor is always the state of the catalog, never the platform.
How does it connect to an ERP?
Using the same principles described in our article on Odoo connectors: message queue, dedicated worker, nightly reconciliation. Nothing magical—just proven engineering.
Conclusion
We didn’t build Pixee Commerce because the market lacked e-commerce platforms. We built it because the protocols that agents expect have become the foundation—and you can’t bolt a foundation onto a finished facade.
To explore the platform: pixeecommerce.com. To discuss your situation without any preconceived notions about the solution, write to us—in half of all cases, our recommendation is to keep the existing platform and fix the product catalog.