What Is the Model Context Protocol (MCP)?
The Model Context Protocol is an open standard that gives AI agents a common way to connect to external data sources, tools and business systems while a conversation is happening. Before MCP, every AI integration was a custom build. A store that wanted its catalog readable by ChatGPT needed one integration, Claude needed another, and each new assistant meant more engineering. MCP replaces those one-off connections with one standard.
Anthropic released MCP in November 2024. OpenAI adopted it in March 2025 and Google DeepMind in April 2025. In December 2025 Anthropic donated the protocol to the Linux Foundation's Agentic AI Foundation, where it is now governed as a vendor-neutral standard with contributions from AWS, Google, Microsoft and OpenAI. Anthropic reported 97 million monthly SDK downloads, more than 10,000 active servers and over 300 client applications at the time of the donation (Anthropic, December 2025).
For what the donation changes for retailers, see Paz.ai's analysis of the MCP donation and Adobe's move to make MCP the default agent protocol for commerce.
Why MCP Matters for Retailers
MCP is how an agent reads live product, inventory and price data instead of relying on whatever it learned in training.
In agentic commerce the protocols divide the work. ACP handles checkout, UCP covers the commerce journey, and MCP is the connection an agent uses to read a merchant's systems: product data, inventory, pricing and orders, as they are right now.
Without a live connection of this kind, an agent only knows what was in its training data or in a feed that was last refreshed hours ago. With one, it can answer "is this jacket in stock in a medium" from the store's own system.
Adoption is broad. Every major AI platform (OpenAI, Google, Anthropic, Microsoft) supports it, and Google designed UCP to work with MCP as one of three integration paths alongside REST APIs and A2A (Google Developers, January 2026). Shopify runs MCP servers that let assistants manage store operations, listings and customer interactions directly (Shopify, 2025).
How MCP Works in Commerce
Agents are MCP clients. Business systems are MCP servers that expose specific capabilities such as catalog search or inventory checks.
MCP uses a client-server model. The AI agent (ChatGPT, Claude, Gemini) is the client. The business system is the server, and it exposes a defined set of capabilities.
A retailer might run an MCP server in front of its product database, inventory system and order system. When a shopper asks an assistant whether a product is available in a given size, the assistant queries that server for the current answer rather than guessing from older data.
The protocol handles authentication through OAuth 2.1 (added in the June 2025 specification update), supports local and remote connections, and standardizes error handling and capability negotiation. A commerce implementation typically exposes capabilities like these:
- Product search: query the catalog by attribute, category or price range
- Inventory check: current stock by size, color or location
- Pricing: current prices including active promotions and loyalty discounts
- Cart operations: add or remove items, calculate totals, apply coupons
- Order status: shipment tracking and delivery estimates
MCP vs ACP vs UCP
MCP is the data layer underneath both ACP and UCP. They complement each other rather than compete.
Scroll horizontally to read the full table →
| Attribute | MCP | ACP | UCP |
|---|---|---|---|
| Primary role | Data connectivity layer | Checkout protocol | Full commerce lifecycle |
| Commerce-specific? | No (general purpose) | Yes | Yes |
| Authentication | OAuth 2.1 | Stripe payment credentials | AP2 payment credentials |
| Maintainer | Anthropic, now Linux Foundation | Stripe and OpenAI | Google and Shopify |
| Adoption | 97M+ monthly SDK downloads | Platform-specific onboarding | 20+ launch partners |
MCP connects applications to tools and data. ACP and UCP describe commerce interactions. Running an MCP server or supporting a commerce protocol does not by itself enroll a merchant on a platform, make products visible in answers or enable checkout. Those remain separate steps with each platform.
What an Ecommerce Team Should Do About MCP
Most stores will meet MCP through their platform, not by building a server. The product information behind the connection is the part worth working on now.
For most brands, MCP arrives through the commerce platform or the tools already in use. Shopify, Adobe Commerce and others are shipping the servers. Building your own is an engineering decision for teams with a specific need, not a prerequisite for showing up in AI shopping.
What every connection, MCP or otherwise, exposes is the same product information: titles, attributes, descriptions, availability and the answers to the questions shoppers ask. If that information is incomplete or inconsistent, a faster pipe does not help. That is the work Paz handles for ecommerce teams: a Paz expert finds the gaps that matter for your products, fixes them in your sources, and tracks how products appear in AI answers over time. Our experts publish within agreed guidelines and bring exceptions to you. See how the product content service works.
FAQ
Do I need to build an MCP server to sell through AI agents?+
Is MCP only for commerce?+
How secure is MCP?+
Can MCP replace my existing APIs?+
Related terms
How AI-ready are your products?
Check one product page for the information gaps that keep it out of AI answers and search results. If you want the fixes handled, Paz experts do that work for ecommerce teams.