But the exam syllabus is public. And the knowledge it tests is exactly what you need to build production-grade AI applications.

A developer named Hooeem broke down the full curriculum — extracting what matters and making it available without the certification gatekeeping.

What the Exam Actually Tests

The Claude Certified Architect certification covers four core areas:

1. Claude Code — CLAUDE.md configuration, plan mode, slash commands, and how to structure agent instructions for consistent, high-quality output.

2. Claude Agent SDK — Building production agent systems. Multi-agent coordination, orchestrator-subagent patterns, tool use, and handling complex, multi-step workflows.

3. Claude API — The full API surface. Structured outputs with JSON schema, tool_use patterns, validation loops, and streaming.

4. Model Context Protocol (MCP) — How to build and consume MCP servers. Connecting agents to real-world data sources, APIs, and tools through the standardized protocol.

The Domain Breakdown

The exam weights these domains:

  • Agent Architecture & Orchestration: 27% — the biggest chunk, covering how agents talk to each other and how to design reliable multi-agent systems
  • Claude Code in Practice: patterns for using Claude Code in CI/CD, non-interactive pipelines, and automated workflows
  • API Mastery: structured data extraction, error handling, rate limiting
  • MCP Integration: building connectors, consuming existing servers, debugging

Why This Matters Without the Cert

You don't need a certification to build with Claude. You need the knowledge the certification tests.

The five project types the exam uses as assessment scenarios are genuinely useful to understand:

  1. Customer support agent (Agent SDK + MCP + escalation handling)
  2. Code generation with Claude Code (CLAUDE.md + planning + slash commands)
  3. Multi-agent research system (coordinator-subagent orchestration)
  4. Developer productivity tooling (built-in tools + MCP servers)
  5. CI/CD integration (non-interactive pipelines + structured output)

If you can build all five of those reasonably well, you can build most production AI applications.

The certification is closed. The knowledge isn't. Work through those five project types and you'll have most of what the exam tests — plus working code.