Episode 6: MCP Protocol - Claude's 'Outsourced Team'

⏱ Est. reading time: 4 min Updated on 5/7/2026

The Model Context Protocol (MCP) is the heart of Claude Code's extensibility. Through MCP, you can connect Claude to almost any external service—from database queries to real-time weather, to advanced image processing.

What is MCP?

Think of it like this:

  • Built-in Tools = Internal employees, available immediately for basic tasks.
  • MCP Servers = Specialized "outsourced teams," each responsible for complex tasks in specific domains.
  • MCP Protocol = Communication protocol, ensuring that reports submitted by outsourcing teams match company standards.
flowchart TB
    subgraph Claude Code
        CC[Claude AI]
        TE[Tool Engine]
    end

    subgraph MCP Protocol Layer
        PROTO[MCP Protocol
JSON-RPC 2.0]
    end

    subgraph MCP Servers
        S1[MySQL MCP]
        S2[Brave Search MCP]
        S3[Image Analysis MCP]
    end

    CC --> TE
    TE <-->|stdio / SSE| PROTO
    PROTO <--> S1
    PROTO <--> S2
    PROTO <--> S3

Two Transport Methods

Method Principle Use Case
stdio Claude Code spawns a child process, communicating via stdin/stdout Most locally installed tools, like MySQL or Git
SSE Connects to remote services via HTTP protocol Centralized cloud services or tools used across devices

Configuration Example (stdio mode)

MCP is typically configured in settings.json or settings.local.json:

"mcpServers": {
  "mysql": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "@anthropic-ai/mcp-server-mysql"],
    "env": {
      "MYSQL_HOST": "localhost",
      "MYSQL_DATABASE": "production"
    }
  }
}

Naming Convention

In a conversation, MCP Tools appear in the format mcp__<server-name>__<tool-name>. For example, the query function provided by the MySQL MCP would be identified as mcp__mysql__query.