Welcome to Part 1 of our exclusive architecture series analyzing Claude Code. As AI assistants transition from chat windows directly into our IDEs and terminals, understanding how they are built at a structural level provides incredible insights for modern software engineering.

A Layered Design

Claude Code is a massively complex TypeScript application holding approximately 1,884 files and generating over 512,000 lines of code. It achieves stability through strict domain separation:

  1. Entry Points: The out-facing interfaces (cli.tsx, mcp.ts).
  2. Bootstrap: Where environment telemetry, proxy setup, and TLS overrides happen before a single frame renders.
  3. Presentation (UI): A terminal-native rendering pipeline built entirely via custom React components processing Ink library constructs.
  4. Query Engine: The event loop governing API execution, context limits, and logical inference state.
  5. Tool System: The payload processing block defining 50+ integrations (like bash, ripgrep, web fetching).
  6. Persistence: Local .jsonl logging and robust Session Memory databases.

Fast-Path Routing & Bootstrapping

In traditional CLI apps, importing large dependencies blocks startup times. Claude circumvents this using Fast-Path Routing.

When a user executes ./claude --help, the command never loads the primary React engine. It evaluates arguments instantly via cli.tsx and halts immediately. Only full interactive requests pass through into parallel Promise.all startup blocks—simultaneously fetching required git configurations, checking anthropic API availability dynamically via TCP pings, and indexing local .claude definitions.

Understanding how Claude constructs these foundational layers sets the stage for our next topic. In Part 2, we will shift focus completely toward its Async Query Engine and how it natively runs thousands of operations via the terminal.