Claude Certified Architect – Foundations Exam Guide

Everything you need to know about the CCA-F certification exam, from preparation strategies to exam day tips.

Exam Information

90 minutes

Duration

60 questions

Questions

720 (scaled)

Passing Score

TBD

Exam Cost

Recommended Study Plan

Week 1

Agentic Architecture & Orchestration

Design and implement agentic systems with Claude

10-12 hours

Agentic loop pattern: send request → inspect stop_reason → execute tools → return results → repeat
stop_reason values: tool_use, end_turn, max_tokens, stop_sequence, refusal, pause_turn
Multi-agent hub-and-spoke architecture: coordinator manages all subagent communication
Subagent isolation: context must be explicitly passed, no automatic inheritance
Task tool for spawning subagents with allowedTools restrictions
Dynamic subagent selection based on query complexity
Iterative refinement loops: evaluate → re-delegate → re-synthesize
Error recovery: descriptive tool_result errors, is_error flag, exponential backoff

Recommended Resources:

Anthropic Docs

Claude Architect Study Notes

Week 2

Tool Design & MCP Integration

Design tools and integrate via the Model Context Protocol

8-10 hours

MCP architecture: client-server protocol with JSON-RPC 2.0 messaging
MCP primitives: Tools (callable functions), Resources (data/context), Prompts (templates)
Transport layers: stdio for local processes, SSE/HTTP for remote servers
Tool schema best practices: descriptive names, clear descriptions, minimal parameters
Error handling in tools: is_error flag, actionable messages, no raw stack traces
MCP server configuration: .mcp.json (project), settings.json (user), enterprise
Security: input validation, least privilege, secrets via environment variables
allowedTools restrictions per agent and MCP server

Recommended Resources:

MCP Specification

Claude Tool Use Guide

Week 3

Claude Code & Prompt Engineering

Configure Claude Code workflows and master prompt engineering

10-12 hours

CLAUDE.md hierarchy: enterprise → user → project root → subdirectory
Custom slash commands: .claude/commands/ directory, $ARGUMENTS placeholder
Hooks: PreToolUse, PostToolUse, Notification — configured in settings.json
Headless mode: claude -p 'prompt' --output-format json --allowedTools
CI/CD patterns: PR review, test generation, documentation, issue triage
System prompts: role, task, constraints with XML tag organization
Response prefilling: force output format by pre-filling assistant turn
Structured output: tool use schemas, prefilling, response_format parameter

Recommended Resources:

Claude Code Docs

Prompt Engineering Guide

Week 4

Context Management & Review

Manage context windows, caching, and ensure system reliability

8-10 hours

Context window optimization: system prompt + history + tools all consume tokens
Prompt caching: cache_control ephemeral markers, 5-min TTL, 90% cost reduction
Minimum cache sizes: 1024 tokens (Sonnet), 2048 tokens (Opus)
Compaction: automatic context compression when approaching window limits
Extended thinking: budget_tokens for complex reasoning, separate thinking block
Evaluation: automated metrics, A/B testing, regression tracking across versions
Production reliability: rate limiting, fallback strategies, monitoring, idempotency
Review all 5 domains, take practice exams, focus on weakest areas
Practice scenario-based questions: customer support agent, code generation, multi-agent research

Recommended Resources:

Prompt Caching Docs

Practice Exams

Exam Tips & Strategies

Focus on Scenarios

The exam uses scenario-based questions from 6 real-world contexts. Understand HOW to apply concepts, not just definitions. Practice identifying which architectural pattern fits each scenario.

Master stop_reason

The agentic loop and stop_reason handling is the foundation of Domain 1 (27%). Know every stop_reason value and the correct action for each. This is the most heavily tested concept.

Time Management

You have 90 minutes for 60 questions — about 90 seconds per question. Scenario-based questions require careful reading. Flag complex questions and return to them after completing easier ones.

Know MCP vs Direct Tool Use

Understand when to use MCP servers versus direct tool definitions. MCP is better for standardized, reusable integrations; direct tool use is simpler for application-specific tools.

Key Topics to Master

Agentic Loop Pattern — stop_reason-based loop control

Multi-Agent Orchestration — coordinator-subagent hub-and-spoke

MCP Protocol — resources, tools, prompts, transports

Tool Schema Design — descriptive names, input_schema, error handling

CLAUDE.md — configuration hierarchy and best practices

Headless Mode — CI/CD integration with --allowedTools

System Prompts — XML tags, role/task/constraints structure

Structured Output — tool use schemas vs prefilling vs response_format

Prompt Caching — cache_control, TTL, minimum token thresholds

Extended Thinking — budget_tokens, when to use, limitations

Recommended Study Plan

Week 1

Agentic Architecture & Orchestration

Design and implement agentic systems with Claude

10-12 hours

Agentic loop pattern: send request → inspect stop_reason → execute tools → return results → repeat
stop_reason values: tool_use, end_turn, max_tokens, stop_sequence, refusal, pause_turn
Multi-agent hub-and-spoke architecture: coordinator manages all subagent communication
Subagent isolation: context must be explicitly passed, no automatic inheritance
Task tool for spawning subagents with allowedTools restrictions
Dynamic subagent selection based on query complexity
Iterative refinement loops: evaluate → re-delegate → re-synthesize
Error recovery: descriptive tool_result errors, is_error flag, exponential backoff

Recommended Resources:

Anthropic Docs

Claude Architect Study Notes

Week 2

Tool Design & MCP Integration

Design tools and integrate via the Model Context Protocol

8-10 hours

MCP architecture: client-server protocol with JSON-RPC 2.0 messaging
MCP primitives: Tools (callable functions), Resources (data/context), Prompts (templates)
Transport layers: stdio for local processes, SSE/HTTP for remote servers
Tool schema best practices: descriptive names, clear descriptions, minimal parameters
Error handling in tools: is_error flag, actionable messages, no raw stack traces
MCP server configuration: .mcp.json (project), settings.json (user), enterprise
Security: input validation, least privilege, secrets via environment variables
allowedTools restrictions per agent and MCP server

Recommended Resources:

MCP Specification

Claude Tool Use Guide

Week 3

Claude Code & Prompt Engineering

Configure Claude Code workflows and master prompt engineering

10-12 hours

CLAUDE.md hierarchy: enterprise → user → project root → subdirectory
Custom slash commands: .claude/commands/ directory, $ARGUMENTS placeholder
Hooks: PreToolUse, PostToolUse, Notification — configured in settings.json
Headless mode: claude -p 'prompt' --output-format json --allowedTools
CI/CD patterns: PR review, test generation, documentation, issue triage
System prompts: role, task, constraints with XML tag organization
Response prefilling: force output format by pre-filling assistant turn
Structured output: tool use schemas, prefilling, response_format parameter

Recommended Resources:

Claude Code Docs

Prompt Engineering Guide

Week 4

Context Management & Review

Manage context windows, caching, and ensure system reliability

8-10 hours

Context window optimization: system prompt + history + tools all consume tokens
Prompt caching: cache_control ephemeral markers, 5-min TTL, 90% cost reduction
Minimum cache sizes: 1024 tokens (Sonnet), 2048 tokens (Opus)
Compaction: automatic context compression when approaching window limits
Extended thinking: budget_tokens for complex reasoning, separate thinking block
Evaluation: automated metrics, A/B testing, regression tracking across versions
Production reliability: rate limiting, fallback strategies, monitoring, idempotency
Review all 5 domains, take practice exams, focus on weakest areas
Practice scenario-based questions: customer support agent, code generation, multi-agent research

Recommended Resources:

Prompt Caching Docs

Practice Exams

Exam Tips & Strategies

Focus on Scenarios

The exam uses scenario-based questions from 6 real-world contexts. Understand HOW to apply concepts, not just definitions. Practice identifying which architectural pattern fits each scenario.

Master stop_reason

The agentic loop and stop_reason handling is the foundation of Domain 1 (27%). Know every stop_reason value and the correct action for each. This is the most heavily tested concept.

Time Management

You have 90 minutes for 60 questions — about 90 seconds per question. Scenario-based questions require careful reading. Flag complex questions and return to them after completing easier ones.

Know MCP vs Direct Tool Use

Understand when to use MCP servers versus direct tool definitions. MCP is better for standardized, reusable integrations; direct tool use is simpler for application-specific tools.

Key Topics to Master

Agentic Loop Pattern — stop_reason-based loop control

Multi-Agent Orchestration — coordinator-subagent hub-and-spoke

MCP Protocol — resources, tools, prompts, transports

Tool Schema Design — descriptive names, input_schema, error handling

CLAUDE.md — configuration hierarchy and best practices

Headless Mode — CI/CD integration with --allowedTools

System Prompts — XML tags, role/task/constraints structure

Structured Output — tool use schemas vs prefilling vs response_format

Prompt Caching — cache_control, TTL, minimum token thresholds

Extended Thinking — budget_tokens, when to use, limitations

Quick Feedback

Claude Certified Architect – Foundations Exam Guide

Exam Information

Recommended Study Plan

Exam Tips & Strategies

Focus on Scenarios

Master stop_reason

Time Management

Know MCP vs Direct Tool Use

Key Topics to Master

Claude Certified Architect – Foundations Exam Guide

Exam Information

Recommended Study Plan

Exam Tips & Strategies

Focus on Scenarios

Master stop_reason

Time Management

Know MCP vs Direct Tool Use

Key Topics to Master