What Is Model Context Protocol (MCP) and How It Empowers LLMs?

What Is MCP

Model Context Protocol (MCP) stands for Model Context Protocol, where “Model” refers to the LLM, “Context” highlights its role as an information hub, and “Protocol” emphasizes standardized interaction. Like a USB‑C interface, MCP provides a unified way for LLMs to connect with external capabilities.

In practice, an LLM (the compute core) may need to access local data (e.g., disk storage) and remote services (e.g., email servers). MCP acts as a converter that integrates these multi‑source abilities, allowing the LLM to go beyond its trained knowledge by invoking external resources through a standard protocol.

Composition

MCP Host : The entity that integrates external capabilities, typically the user‑facing interface such as an LLM desktop app or a productivity tool built on an LLM. It coordinates interactions among the user, the LLM, and external resources.

MCP Client : A component created by the Host that maintains a stateful session with the MCP Server, translating Host requests into MCP‑compliant messages and parsing Server responses.

MCP Server : The concrete implementation that exposes external abilities (e.g., reading files, querying weather) via a standardized protocol. It does not contain the resources itself but acts as a proxy.

Server‑Provided Abilities

Resources : Static or semi‑dynamic data that the LLM can ingest directly, such as logs or configuration files.

Tools : Executable actions that fulfill specific sub‑tasks, like writing to a database or invoking a third‑party system.

Prompts : Reusable LLM prompt templates that guide the model’s behavior for particular scenarios.

Analogously, Resources are raw ingredients, Tools are the kitchen utensils, and Prompts encode the chef’s preferences.

Interactive Demo

Using the “Tongyi Lingma” VS Code extension (which embeds an MCP Client), a user can register an external MCP Server for 12306 ticket queries. After switching the extension to “Agent” mode, the user adds the “12306‑MCP” service, enabling the LLM to call four MCP tools sequentially: retrieve current time, get departure station code, get destination station code, and fetch ticket information. The LLM then selects the fastest train based on the gathered data.

Even when asking for tickets with second‑class seats, the LLM can filter results itself if the Server does not provide a dedicated tool.

Building a Simple MCP Server

Developers can quickly create a minimal MCP Server in Node.js, Python, or Java. The article shows a simple Python example defining a “1+1=3” tool and demonstrates how the MCP Client can invoke it, illustrating how custom business logic can be exposed to LLMs.

Why MCP Matters

While OpenAI’s Function Calling provides similar capabilities, it is model‑specific and requires different formats for each vendor. MCP offers a vendor‑agnostic, standardized protocol that reduces integration cost and accelerates innovation. By lowering barriers, more business teams can expose their knowledge bases, manuals, and operational tools to LLMs, enhancing productivity and reducing support overhead.

Adopting MCP also raises challenges such as tool composition, context management, authentication, and protocol security, but these are typical of emerging technologies.