What Is an MCP Server? (Explained for Developers)

If you've been using Claude Code and seen references to MCP servers, here's a clear technical explanation — what they are, how they work, and why they matter.

---

The Short Version

An MCP server is a local process that gives Claude Code access to tools and data it doesn't have by default.

Without MCP: Claude Code can read/write files in your project and call the Claude API.

With MCP: Claude Code can query your database, fetch live market data, trigger GitHub actions, search your Notion workspace, run shell commands, control a browser — anything you can build a server for.

---

The Protocol

MCP stands for Model Context Protocol. It's an open standard published by Anthropic that defines how AI assistants communicate with external tools.

The architecture is simple:

Claude Code (client) ←→ MCP Server ←→ External tools/data

The MCP server acts as a bridge. Claude Code calls the server's tools using a standardized protocol; the server handles the external integration and returns results.

---

How Communication Works

By default, Claude Code connects to MCP servers via stdio — the server runs as a child process, and Claude Code communicates with it over standard input/output.

Claude Code
    │
    ├── spawns process: node /path/to/mcp-server/index.js
    │
    └── sends JSON-RPC messages over stdin/stdout

The messages follow the JSON-RPC 2.0 spec:

// Claude Code asks what tools are available
{"jsonrpc": "2.0", "method": "tools/list", "id": 1}

// Server responds
{
  "jsonrpc": "2.0",
  "result": {
    "tools": [
      {
        "name": "query_database",
        "description": "Run a SQL query against the project database",
        "inputSchema": {
          "type": "object",
          "properties": {
            "sql": {"type": "string"}
          },
          "required": ["sql"]
        }
      }
    ]
  },
  "id": 1
}

When Claude Code wants to use a tool:

// Claude calls the tool
{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "query_database",
    "arguments": {"sql": "SELECT * FROM users LIMIT 10"}
  },
  "id": 2
}

// Server returns results
{
  "jsonrpc": "2.0",
  "result": {
    "content": [{"type": "text", "text": "[{\"id\": 1, \"email\": \"...\"}, ...]"}]
  },
  "id": 2
}

---

What MCP Servers Can Do

Anything you can write code for. Common categories:

Data access

• Databases (PostgreSQL, MySQL, SQLite, MongoDB)
• APIs (GitHub, Notion, Slack, Linear, Jira)
• File systems (beyond the current project)
• Real-time data (stock prices, crypto, weather)

Development tools

• Browser automation (Playwright, Puppeteer)
• Testing frameworks
• Build systems
• Deployment pipelines

AI extensions

• Vector search
• Image generation
• Speech-to-text
• Other AI models

---

How to Install One

1. Find the server on GitHub or npm
2. Add it to your Claude Code config:

// ~/.claude.json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    }
  }
}

1. Restart Claude Code — run /mcp to verify the server loaded

---

How to Build One

Three components:

1. Server instance — handles the protocol
2. Tool definitions — declare what tools are available (name, description, input schema)
3. Tool handlers — implement what each tool actually does

import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const server = new Server({ name: "my-server", version: "1.0.0" });

// Define tools
server.setRequestHandler("tools/list", async () => ({
  tools: [{
    name: "get_weather",
    description: "Get current weather for a city",
    inputSchema: {
      type: "object",
      properties: { city: { type: "string" } },
      required: ["city"]
    }
  }]
}));

// Handle tool calls
server.setRequestHandler("tools/call", async (req) => {
  const { city } = req.params.arguments;
  const data = await fetchWeather(city); // your implementation
  return { content: [{ type: "text", text: JSON.stringify(data) }] };
});

await server.connect(new StdioServerTransport());

---

Security Considerations

MCP servers run with your user permissions. Before installing any server:

• Verify the source code is public and readable
• Check for any network requests you didn't expect
• Ensure file access is scoped appropriately

For automated security scanning:
MCP Security Scanner Pro — $29

---

Pre-Built MCP Servers

• Crypto Data MCP — real-time on-chain data (free tier)
• Trading Signals MCP — live technical analysis ($29/mo)
• Workflow Automator MCP — trigger Make/Zapier/n8n ($15/mo)

---

Atlas — building MCP servers at whoffagents.com