Claude Code & MCP

You have built an agent from scratch. Production teams rarely start from scratch — they use agentic coding tools like Claude Code, and they connect those agents to real systems with the Model Context Protocol (MCP). This lesson shows where your hand-built understanding meets the tools you will actually use.

Overview

Claude Code is an agentic coding tool: it runs in your terminal, reads and edits files, runs commands, and works through multi-step tasks using exactly the agent loop you built last week — just hardened and productised. MCP is an open standard for connecting such agents to external tools and data, so you do not write a bespoke integration for every system.

Key ideas

Agentic coding versus a chat box

A chat box gives you text you copy back into your editor. An agentic tool acts directly: it edits files, runs tests, reads errors, and iterates. The model is the same kind you have been calling; the difference is the harness around it — file access, command execution, and a loop with stopping conditions, all wired together.

What MCP standardises

Before MCP, every tool integration was custom: one way to connect a model to your database, another for your file store, another for an API. MCP defines a single protocol so any MCP-compatible client (like Claude Code) can talk to any MCP server (a wrapper around a tool or data source). Write the server once, and every compatible agent can use it.

   Agent (MCP client)              Your systems (MCP servers)
   ------------------              --------------------------
   Claude Code        <-- MCP -->  database server
                      <-- MCP -->  filesystem server
                      <-- MCP -->  your custom API server

Connecting a server

MCP servers are registered with the client through configuration. Conceptually you point the client at a server command or endpoint, and its tools become available to the agent — the same name, description, and input schema mechanism from the tool-use lesson, just discovered over a protocol instead of hardcoded in your script.

{
  "mcpServers": {
    "my-database": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgres://localhost/shop"]
    }
  }
}

Why this matters for shipping

When you build a real product, you want the agent to reach your actual systems safely and consistently. MCP gives you a reusable, inspectable boundary: each server declares exactly which tools it exposes, so you control the surface area. It is the production-grade version of the tools list you wrote by hand.

Quick recap

• Claude Code is an agentic coding tool: the agent loop you built, hardened for real work.
• MCP is an open standard for connecting agents to tools and data sources.
• Write an MCP server once; any compatible client can use it.
• Servers declare their tools, giving you a controlled, inspectable surface area.
• The tool-use rules from Week 3 still apply — MCP just standardises distribution.