Model Context Protocol (MCP)

Anthropic introduced MCP in November 2024 as an open, open-source standard, and it was adopted broadly across the industry over the following year, including by OpenAI and Google. In December 2025 Anthropic donated the protocol to the Agentic AI Foundation under the Linux Foundation, cementing it as vendor-neutral. It is often described as a USB-C port for AI: a single interface any host and any tool can implement to interoperate.

MCP is the tool and context layer of an agent. It uses a client-server architecture over JSON-RPC: an AI application (the host, such as Claude, ChatGPT, or an IDE assistant) runs MCP clients that connect to MCP servers, which expose three primitives, tools (functions the agent can call), resources (data the agent can read), and prompts (reusable templates). This is how a single agent discovers what it can read and what actions it can take against an external system.

In the agentic-commerce stack, MCP is explicitly NOT a payments or commerce protocol. It is the layer that lets an agent reach the other layers: it standardizes how an agent reads data and invokes tools and APIs, while checkout standards like the Agentic Commerce Protocol (ACP) handle how an agent buys and payment standards like Agent Payments Protocol (AP2) handle how a purchase is authorized. The pieces are designed to compose, and one buying flow can use several.

As commerce agents adopt MCP-style tool access, the way your store becomes reachable by an agent is through machine-readable, structured product data and clean APIs. An agent connected over MCP can only surface, compare, and act on catalog data that is accurate and exposed in a form it can read; a store whose product information is trapped in unstructured pages is effectively invisible to that agent.

This is upstream of any checkout. Before an agent can transact, it has to find and trust your products, which means the durable merchant investment is agent-readiness: accurate structured data (JSON-LD), a clean product feed, and strong AI visibility, so whichever tool or checkout standard an agent uses, your catalog is legible to it.

Illustrative scenario: a shopping assistant connected to a merchant tool over MCP reads product resources (price, availability, variants) and calls a tool to check stock before recommending an item. If the merchant exposes accurate structured data, the agent can surface and reason about the product; if key fields are missing or unstructured, the agent skips it in favor of a competitor it can actually read.