Overview
Claude Code Decoded is a daily series that goes through Claude Code from the inside out — every file, every tool, every system. Episode 1 sets the architectural foundation: the 6 major components, how they fit together, and the 20-episode roadmap for the series.
If you use Claude Code daily as a developer or cloud architect, this series teaches you what is actually happening under the hood — and why that knowledge makes you significantly more effective when using, configuring, and extending the tool.
The 6 Major Components
1. CLI Layer
The command-line interface is the entry point — it handles argument parsing, configuration loading, and session initialisation. Claude Code is a Node.js binary, which means it runs on any platform with Node installed and can be inspected like any JavaScript application.
2. Agentic Loop
The agentic loop is the heart of Claude Code. It is the engine that takes a user message, sends it to the model, processes the response, executes tool calls, feeds results back to the model, and continues until the task is complete or the model yields control back to the user. Every serious agent system — not just Claude Code — is built around some version of this loop.
3. Tool System
Every capability in Claude Code is implemented as a tool: reading files, writing files, executing bash commands, searching codebases, making web requests. The model does not have direct access to your filesystem — it requests tool calls, and the tool system executes them with the appropriate permissions. This architecture makes Claude Code safe to extend and straightforward to audit.
4. MCP (Model Context Protocol)
MCP is the extension system that allows Claude Code to connect to external tools, APIs, and data sources. MCP servers are separate processes that expose additional tools to the Claude Code tool registry. This is how you connect Claude Code to your database, your internal APIs, your observability platform, or any other system.
5. Hooks System
Hooks let you intercept and customise Claude Code behaviour at specific lifecycle points — before a tool executes, after a response is generated, when a session starts or ends. This is the customisation layer for enterprise deployments where you need to enforce policies, log activity, or integrate with existing workflows.
6. Skills and Plugins
Skills are packaged, reusable capabilities that can be loaded into Claude Code sessions. They combine CLAUDE.md instructions, tool configurations, and MCP server definitions into a single deployable unit. This is how organisations can standardise Claude Code behaviour across development teams.
The Agentic Loop Pattern
The agentic loop is the architectural pattern that every serious AI agent system uses. Understanding it at the code level — how Claude Code implements message passing, tool execution, result injection, and loop continuation — transfers directly to building your own agent systems. It is the most important component in the entire series.
The 20-Episode Roadmap
The series covers: CLI entry point, agentic loop internals, built-in tools, MCP architecture, hooks system, skills and plugins, session management, permission system, configuration cascade, and enterprise deployment patterns. Each episode opens the actual source code and explains what it does and why.
Key Takeaways
- Claude Code has 6 major architectural components: CLI, agentic loop, tools, MCP, hooks, and skills
- The agentic loop is the core pattern — understanding it applies to all agent systems, not just Claude Code
- Every capability is a tool — this architecture makes Claude Code auditable and extensible
- MCP is the extension layer for connecting Claude Code to any external system
- The series goes file-by-file through the source code across 20 episodes
