Published on

Claude Certified Architect – Foundations: The Complete Exam Guide

Authors
  • avatar
    Name
    Vuk Dukic
    Twitter

    Founder, AI/ML Engineer

Certified Claude Architect

What Is the Claude Certified Architect – Foundations (CCA-F)?

The Claude Certified Architect – Foundations (CCA-F) is Anthropic's official professional certification for practitioners who design and deploy production-scale Claude applications — particularly agent-based and code-centric systems. It validates your ability to architect reliable, scalable AI workflows using Claude's full ecosystem, including the Claude API, Claude Code, and the Model Context Protocol (MCP).

Exam At a Glance

| Detail | Info | |---|---| | Full Name | Claude Certified Architect – Foundations (CCA-F) | | Administered by | Anthropic (online, proctored) | | Format | 60 scenario-based multiple-choice questions | | Duration | 120 minutes | | Passing Score | 720 out of 1,000 | | Target Audience | Solution/AI architects, developers with 6+ months of hands-on Claude experience | | Pricing | ~$99 per attempt (free for Anthropic Partner Network members) | | Access | Anthropic Skilljar portal |

The Five Exam Domains

Domain 1 — Agentic Architecture & Orchestration (27%)

This is the largest domain and the backbone of the exam. It tests your ability to design and manage agentic loops, multi-agent systems, and complex orchestration patterns.

Key topics:

  • Agentic loop lifecycle: sending requests, inspecting stop_reason (tool_use vs end_turn), executing tools, and returning results
  • Model-driven decisions vs. pre-configured decision trees
  • Hub-and-spoke coordinator–subagent architecture: a central coordinator manages all sub-agent communication, error handling, task decomposition, and result aggregation
  • Sub-agents operating with isolated context (not inheriting the coordinator's full history)
  • Multi-step workflow enforcement: programmatic hooks, prerequisite gates, and lifecycle callbacks
  • Task decomposition: fixed sequential pipelines vs. adaptive dynamic decomposition
  • Investigation plans that generate new subtasks as the agent discovers information

Domain 2 — Tool Design & MCP Integration (18%)

Focuses on how tools are designed, secured, and integrated via the Model Context Protocol (MCP).

Key topics:

  • Defining tools with clear boundaries, resources, and validations
  • Writing precise tool descriptions that prevent misrouting and ambiguity
  • When to split vs. consolidate tools (purpose-specific interfaces vs. monolithic tools)
  • Scoped tool access: limiting agents to role-relevant tools (least-privilege principles)
  • Structured error responses: isError flag, error categories (transient, validation, business, permission)
  • isRetryable semantics and retry-decision logic
  • Ensuring tool results are appended correctly into conversation history

Domain 3 — Claude Code Configuration & Workflows (20%)

Tests your ability to configure and automate Claude Code-based workflows in CI/CD environments.

Key topics:

  • CLAUDE.md configuration and Agent Skills
  • Plan mode and slash commands (/plan, /tools, /context)
  • Pre- and post-run commands
  • Integrating Claude Code into CI/CD pipelines (testing, promoting, versioning skills and agents)
  • Multi-workspace workflows and command bundling
  • Blending editor-driven and agent-driven steps in production environments

Domain 4 — Prompt Engineering & Structured Output (20%)

Covers designing prompts and schemas that reliably drive structured, machine-consumable outputs.

Key topics:

  • Context engineering: role-setting, step-by-step instructions, constraints, and guardrails
  • JSON-schema-based structured output using tool_use-style responses
  • Few-shot prompting, extraction patterns, and explicit criteria
  • Validation-retry loops
  • Batch API usage for high-volume structured extraction
  • Idempotent, deterministic output formats for downstream processing

Domain 5 — Context Management & Reliability (15%)

Focuses on managing long-context constraints, multi-agent handoffs, and error propagation in production systems.

Key topics:

  • Long-context handling: summarization, truncation, and selective injection strategies
  • Multi-agent handoffs: preserving context integrity when switching agents or roles
  • Escalation triggers: user requests for humans, policy gaps, inability to make meaningful progress
  • Ambiguity resolution patterns and escalation to human review
  • Confidence calibration, retry strategies, and monitoring
  • Contextual safeguards to prevent lossy state or circular loops

Exam Scenarios

All 60 questions are scenario-based — no trivia-style fact recall. During the exam, you'll encounter 4 randomly selected scenarios from a pool of 6:

  1. Customer Support Agent
  2. Code Generation with Claude Code
  3. Multi-Agent Research System
  4. Developer Productivity Assistant
  5. Claude Code for CI/CD
  6. Structured Data Extraction

Each scenario tests multiple domains simultaneously, with the heaviest emphasis on Agentic Architecture & Orchestration, Tool Design & MCP, and Prompt Engineering & Structured Output.

How to Prepare

Recommended Study Path

For practitioners with 6+ months of Claude experience: ~15–20 hours of focused study For developers new to Claude: ~30–40 hours, including hands-on labs

Free Resources

  • claudecertificationguide.com — Community-run guide with 30 lessons, 150+ practice questions, and a full mock exam
  • Panaversity CCA-F — 13 free courses explicitly mapped to CCA-F domains
  • Vizuara AI Pods — Free hands-on notebooks covering all five domains plus a 60-question practice exam
  • Anthropic's official practice exam — Available via the Skilljar portal

Paid Resources

  • ExamPro — "Claude Architect Foundations" course aligned to 2025 exam domains
  • CertSafari — 600+ exam-style questions in 60-question blocks

Domain Priority Strategy

If you're short on time, focus in this order:

  1. Domain 1 (27%) — Largest domain; underpins how agents and tools interact
  2. Domain 3 + Domain 4 (40% combined) — Claude Code and Prompt Engineering together form the biggest combined weight
  3. Domain 2 (18%) — Direct hands-on knowledge of MCP servers, tool shapes, and error patterns
  4. Domain 5 (15%) — Context management and reliability patterns

Final Thoughts

The CCA-F is a rigorous, scenario-driven exam that rewards practitioners who have actually built production Claude applications. It's not a memorization test — it's a design test. The best preparation is hands-on experience building agentic workflows, configuring Claude Code, and designing MCP-compliant tools.

If you're an AI architect, solutions engineer, or senior developer working with Claude, this certification is a strong signal of your ability to design reliable, production-grade AI systems.

Ready to get started? Request access at anthropic.skilljar.com.