What is MCP?

Overview

The Model Context Protocol (MCP) is an open standard for connecting AI applications to tools and data. Before MCP, every AI app wrote its own integration for every system, such as GitHub, a database, or a file system. With MCP, a system is wrapped once as an MCP server, and any MCP-capable client (a chat app, an IDE, or an agent) can discover and use it. It is often compared to USB-C for AI integrations.

MCP uses a client-server model over JSON-RPC. The host application (for example an IDE) runs one MCP client per server connection. Servers expose three kinds of capabilities. Tools are actions the model can call, resources are data the app can read, and prompts are reusable templates. Servers run locally over standard input and output, or remotely over HTTP.

When to use it

• Giving an AI assistant access to internal systems in a reusable way.
• Building tools once and using them across several AI clients.
• Standardizing how agents discover and call capabilities.

Where it shows up in interviews

Where it is used in real software

Key terms

Host

The AI application the user interacts with, such as an IDE or chat app.

MCP client

The connector inside the host that talks to one MCP server.

MCP server

A program that exposes tools, resources, and prompts for a system.

Tool

A callable action with a name, description, and JSON input schema.

Transport

How messages travel, such as standard input and output locally or streamable HTTP remotely.

How it works, step by step

1. 1
   Connect

   The host starts or connects to a server and the two negotiate protocol version and capabilities.

2. 2
   Discover

   The client lists the server's tools, resources, and prompts.

3. 3
   Offer to the model

   The host includes tool descriptions in the model's context.

4. 4
   Call

   When the model chooses a tool, the client sends a tools/call request with JSON arguments, often after user approval.

5. 5
   Return results

   The server runs the action and returns content that the host feeds back to the model.

Integrations with and without MCP

4 AI apps each need 5 integrations.

MCP turns an N x M integration problem into N + M.

Implementation

1import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";2import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";3import { z } from "zod";4 5const server = new McpServer({ name: "orders", version: "1.0.0" });6 7server.tool(8  "get_order",9  "Look up an order by ID",10  { id: z.string().describe("Order ID") },11  async ({ id }) => {12    const order = await ordersDb.find(id);13    return { content: [{ type: "text", text: JSON.stringify(order) }] };14  },15);16 17await server.connect(new StdioServerTransport());

Complexity and performance

Trade-offs

Variants and related techniques

Common mistakes

• Installing untrusted MCP servers.

  Fix: Treat servers like dependencies with code access; review, pin versions, and run with least privilege.

• Vague tool descriptions.

  Fix: Write clear names, descriptions, and schemas, since the model chooses tools from them.

• Auto-approving destructive tools.

  Fix: Require user confirmation for writes, deletes, and payments.

Interview questions

Practice problems