Model Context Protocol (MCP)

MCP · Model Context Protocol

Definition

MCP is Anthropic's open standard for connecting AI models to external tools and data uniformly over JSON-RPC; the foundation layer for agentic applications, commerce included.

What MCP is and why it appeared

Before MCP, every AI system needed a bespoke integration with every external tool: its own
documentation, its own authentication scheme, its own data formats. Ten tools and five models meant
fifty integrations.

MCP introduces one protocol: any tool that implements an MCP server becomes automatically available
to any agent with an MCP client. The metaphor the protocol is described with is USB-C for AI — one
standard instead of a drawer of adapters.

How MCP is structured

The protocol runs over JSON-RPC 2.0 and defines three kinds of primitive:

Primitive	What it provides	Example
Tools	Actions the agent can invoke	Place an order, search the catalogue
Resources	Data the agent can read	A product record, an order history
Prompts	Instruction templates	A dialogue scenario

An MCP server may expose any combination of these.

MCP in e-commerce

For a retailer, MCP means building one MCP server for the catalogue and the ordering system and
gaining compatibility with every agent that supports the standard. That lowers the barrier to
agentic commerce considerably.

Practical scenarios:

The store’s own AI assistant reaches the current catalogue, prices and stock through MCP
External buyer or partner agents interact with the store over a standard protocol
An orchestrator coordinating several agents reaches different tools — warehouse, CRM,
recommendations — through one interface

Important: MCP is infrastructure, not a commerce standard for completing purchases.
Transaction security, authorisation and orchestration are solved by application protocols running
above it.

The status of the standard

Anthropic introduced MCP in November 2024. In December 2025 the standard moved to the Agentic AI
Foundation under the Linux Foundation, with OpenAI, Block and other industry participants among the
co-founders. That marks the shift from a vendor tool to an open standard with independent
governance.

Frequently asked questions

What is an MCP server?

An MCP server is any service that implements the protocol and exposes its capabilities to an AI agent. It can be a product database, a CRM, a payment system or internal documentation. The agent addresses every server the same way, over JSON-RPC.

How does MCP differ from function calling?

Function calling is a mechanism inside a specific model for invoking functions defined in the prompt. MCP is a higher-level standard for connecting to external systems independently of the model. Function calling can be one way of implementing an MCP client.

How does MCP relate to commerce protocols such as ACP?

MCP is the transport layer — how an AI connects to a service. ACP and its peers are application protocols for purchases and agent interaction. MCP is analogous to TCP/IP, and the commerce protocols to HTTP running over it.