Claude Certified Architect — Foundations
255 practice questions
Last reviewed: April 2026
Personal notes and resource links for your study journey
Filter by Certification
The Claude Certified Architect — Foundations (CCA-F) is the first AI-company-specific architect certification, launched by Anthropic in March 2026. It validates the ability to design, build, and operate production systems on the Claude platform — spanning the Claude API, Claude Code, and the Model Context Protocol (MCP). The exam targets AI engineers, solutions architects, and developers who build agentic workflows, integrate external tools through MCP servers, and ship reliable Claude-powered applications. At 60 multiple-choice questions in 120 minutes with a 720/1000 passing score and a $99 fee, it sits at a foundational level but demands hands-on familiarity with prompt engineering, tool design, context management, and orchestration patterns unique to the Claude ecosystem.
The heaviest domain at 27%. Covers multi-step agent design, orchestration patterns (sequential, parallel, router), handoff protocols between sub-agents, human-in-the-loop checkpoints, and failure-recovery strategies. Expect questions on when to use a single-agent loop versus a multi-agent graph and how to manage shared state across agent steps.
Weighted at 20%. Tests knowledge of Claude Code setup — CLAUDE.md conventions, permission models, hook configuration, slash commands, MCP server integration within Claude Code, and CI/CD workflows that leverage Claude Code for automated coding tasks. Candidates who have only used the API without Claude Code will find this domain challenging.
Also 20%. Covers system-prompt design, prefilling, chain-of-thought elicitation, extended thinking, structured JSON/XML output via tool_use and constrained generation, and techniques for reducing hallucination. Goes beyond basic prompting into production-grade patterns — prompt versioning, A/B testing, and evaluation frameworks.
Weighted at 18%. Focuses on designing tool schemas, MCP server implementation (resources, tools, prompts, sampling), transport layers (stdio, SSE, Streamable HTTP), security boundaries for tool execution, and best practices for tool descriptions that minimize misuse. Expect scenario-based questions on debugging tool-call failures and handling partial tool results.
The lightest domain at 15%, but critical for production systems. Covers context-window budgeting, summarization strategies, caching (prompt caching, ephemeral caching), retrieval-augmented generation patterns, rate-limit handling, retry/backoff logic, and monitoring/observability for Claude API calls. Questions often pair context-window math with cost-optimization trade-offs.
$130k–$185k–$280k USD annual
Range covers US-based AI engineering and solutions-architect roles where Claude API or LLM integration is a primary skill. Entry-level and non-coastal positions trend toward the low end; senior AI architects at frontier-AI companies and well-funded startups push well above the high end ($300k–$450k+ TC). The certification is new and does not yet carry the hiring weight of established cloud certs — its value is strongest as a signal of hands-on Claude platform expertise paired with a portfolio of shipped AI applications.
Source: levels.fyi 2025–2026 AI/ML engineering data, Glassdoor 2025–2026, LinkedIn salary insights for Claude/LLM-related roles. Figures are approximate; actual compensation depends on role, region, and experience.
Demand for engineers who can build production-grade LLM applications has surged since 2024, with "Claude API," "MCP," and "agentic workflows" appearing in a rapidly growing share of AI-engineering job postings. As the first vendor-issued certification for Claude-specific skills, CCA-F occupies a unique niche: it lets candidates demonstrate familiarity with the Claude ecosystem — prompt engineering, tool use, Claude Code, and MCP — in a way no generic AI cert covers. Early adoption is strongest among AI consultancies, startups building on the Anthropic stack, and enterprise teams standardizing on Claude for internal tooling. The certification is most valuable when paired with demonstrable project work (open-source MCP servers, shipped Claude-powered products); on its own it signals foundational competence rather than deep expertise.
There are no formal prerequisites. Anthropic recommends at least three months of hands-on experience building with the Claude API, including practical exposure to tool use, structured output, and at least one agentic workflow that went beyond a single prompt-response cycle.
Familiarity with the Model Context Protocol (MCP) and Claude Code is strongly recommended. Several questions assume you can reason about MCP server architecture (resources, tools, prompts, transport layers) and Claude Code configuration (CLAUDE.md, hooks, permissions). Candidates who have only used Claude through the chat interface without API or CLI experience will find the exam significantly harder than the "foundations" label suggests.
CCA-F is positioned as a foundational certification but should not be mistaken for an easy introductory exam. The 60-question, 120-minute format gives a comfortable two minutes per question, and all questions are multiple-choice — there are no labs or performance-based tasks. However, questions frequently present realistic architectural scenarios that require combining knowledge across domains: for example, choosing an orchestration pattern while also considering context-window limits and tool-call reliability.
Common stumbling blocks include the Claude Code domain (unfamiliar to API-only users), context-window budget calculations with prompt caching, MCP transport-layer trade-offs, and questions that test the boundary between Claude API features and general LLM concepts. Plan on 15–25 hours of study if you actively build with Claude daily, 40+ hours otherwise. The $99 fee and online proctoring make retakes relatively painless.
Inaugural version. 60 MCQ, 120 minutes, 720/1000 pass, $99 USD, online-proctored. Covers Claude API (including extended thinking and tool use), Claude Code, MCP, agentic orchestration, prompt engineering, and context management. Two-year validity.
CCA-F (Claude Certified Architect — Foundations) is a considered an entry-level exam testing breadth of conceptual understanding rather than hands-on depth Foundational-level exam. Most candidates need 30–80 hours of study spread over 3–6 weeks for foundational-level exams. Most candidates who score consistently above the passing threshold on practice exams pass on their first attempt.
Most candidates need 30–80 hours of study spread over 3–6 weeks for foundational-level exams. Time-to-pass varies widely by prior experience. Engineers with hands-on production experience in the underlying technology typically need less; candidates new to the platform should plan toward the upper end of that range.
CCA-F is a recognized credential in the Anthropic ecosystem and signals validated knowledge to employers, recruiters, and clients. Whether it is worth the time and fee for you depends on your role and goals — it tends to pay off most for cloud engineers, architects, and consultants who work with Anthropic day-to-day or want to move into roles that do.
The passing score for CCA-F is 720 / 1000. The exam contains 60 questions and lasts 2 hr.
The CCA-F exam fee is $99 USD. Fees are set by Anthropic and may vary by region; always confirm the current price on the official Anthropic certification page before booking.
Anthropic has not yet published an explicit validity period for the CCA-F certification. The program launched in March 2026; renewal policies are expected to be announced as it matures.
Yes, Anthropic certifications are delivered online only — there are no in-person test centers. The exam runs in a secure proctored browser; you'll need a quiet private room, webcam, microphone, stable broadband, and a government photo ID.
CertLabPro provides 15 study modes across the practice question bank for CCA-F. The exam-simulation mode mirrors the real exam: 60 questions in 2 hr, with the same passing threshold of 720 / 1000. Browse mode lets you read every Q&A statically.