The Claude Certified Architect (CCA) Foundations exam tests you across five domains, each carrying a different percentage of your final score. Knowing the weight of each domain is the first step to studying efficiently — don't spend equal time on a 15% domain and a 27% domain.
Here is the complete breakdown.
Domain 1 — Agentic Architecture (27%)
The largest domain on the exam. It covers how Claude operates inside autonomous loops: tool calling, multi-agent orchestration, subagent design, and reliability patterns like retries, checkpointing, and human-in-the-loop interrupts.
Key topics to master:
- The agent loop — observe, think, act, respond
- Orchestrator vs. subagent roles
- When to use parallel vs. sequential agents
- Scoped tool access and reversible action principles
- Handling ambiguity before long autonomous runs
Domain 2 — Claude Code Configuration (20%)
Covers the configuration layer for Claude Code: CLAUDE.md files, project-level vs. global settings, slash commands, MCP server integration, and memory files.
Key topics to master:
- CLAUDE.md structure and inheritance hierarchy
- When to use
settings.jsonvs.settings.local.json - Defining and invoking custom slash commands
- Connecting MCP servers via
mcpServersconfig - Bash and file tool permission allowlists
Domain 3 — Prompt Engineering (20%)
Tests your ability to write effective system prompts and user turns, and to diagnose why a prompt underperforms.
Key topics to master:
- Persona and role framing, explicit instructions, and context construction
- Few-shot prompting and example selection
- Chain-of-thought and extended thinking usage
- Avoiding prompt injection in production systems
- Structured output via tool_use and JSON schemas
Domain 4 — Tool Design & MCP (18%)
Covers the Model Context Protocol — how to design tool schemas, decide when to split versus consolidate tools, and secure MCP servers in production.
Key topics to master:
- MCP server architecture (resources, tools, prompts)
- JSON Schema for tool input definitions
- Splitting vs. consolidating tools for clarity and reliability
- Tool naming conventions and description writing for LLM use
- Authentication and authorization in MCP servers
Domain 5 — Context Management (15%)
The smallest domain, but one where candidates often leave easy points on the table. It tests knowledge of token budgets, caching strategies, and how to design multi-turn conversations that don't degrade over time.
Key topics to master:
- Progressive summarization and lost-in-the-middle effects
- Prompt caching for reducing token cost and latency
- Conversation compaction and summarisation patterns
- Context window sizing and token estimation
- Multi-turn state management in agentic systems
How to allocate your study time
A simple way to weight your preparation is to match it to the exam distribution:
| Domain | Weight | Suggested study share |
|---|---|---|
| Agentic Architecture | 27% | ~3 hours |
| Claude Code Config | 20% | ~2 hours |
| Prompt Engineering | 20% | ~2 hours |
| Tool Design & MCP | 18% | ~2 hours |
| Context Management | 15% | ~1.5 hours |
After one pass through the material, use the practice questions to identify which domains you're weakest in, then re-allocate time accordingly. The full timed exam simulation is the best final check — it shows you your score broken down by domain so you know exactly where to focus in the last 24 hours before your test.