The Claude Certified Architect – Foundations (CCA-F) is Anthropic’s first official technical certification for practitioners who design and deploy production-grade Claude applications. Launched on March 12, 2026, it is a proctored, closed-book exam focused on real architectural decisions rather than general AI awareness. The exam consists of 60 scenario-based questions, must be completed in 120 minutes, and requires a scaled passing score of 720 out of 1,000. It tests practical expertise across agentic workflows, multi-agent orchestration, Model Context Protocol (MCP) integration, context management, structured output, and Claude Code-based development. To earn the CCA-F credential, candidates need hands-on production experience with the Claude API, the Agent SDK, and Claude Code. This is an architecture-level exam, not a beginner’s credential.
CCA-F Quick Facts
| Detail | Specification |
| Full name | Claude Certified Architect – Foundations |
| Acronym | CCA-F |
| Launch date | March 12, 2026 |
| Questions | 60 multiple-choice |
| Duration | 120 minutes |
| Level | 301 |
| Passing score | 720 / 1,000 (scaled) |
| Format | Scenario-based; 4 of 6 scenarios selected randomly |
| Retakes | One attempt per registration period |
| Price | $125 USD |
| Validity | 12 months |
| Access | Claude Partner Network only (as of June 2026) |
| Badge | Digital CCA-F credential upon passing |
What Is the Claude Certified Architect – Foundations (CCA-F)?
The Claude Certified Architect – Foundations certification is Anthropic’s first official technical credential. It validates a practitioner’s ability to design, orchestrate, and deploy production-grade Claude applications—not just use Claude through a chat interface.
The exam is closed-book, proctored, and entirely scenario-based. Questions place candidates in realistic production environments and test architectural decisions. Distractors reflect common real-world mistakes experienced engineers make.
This ~301-level exam assumes knowledge of prompt writing. It evaluates whether you can build reliable agentic systems under production pressures: tool failures, context overflow, subagent isolation, and escalation edge cases.
CCA-F is the entry point to Anthropic’s planned multi-tier credential stack (additional developer, seller, and advanced architect certifications coming later in 2026). Passers get priority access to future tiers.
Who Should Pursue the Claude Certified Architect Certification?
The official target profile is a solution architect with at least six months of hands-on production experience with the Claude API, Claude Code, the Claude Agent SDK, and the Model Context Protocol (MCP).
You are a strong fit if you:
- Design agentic systems using the Claude Agent SDK with multi-agent orchestration
- Configure Claude Code for teams, including CLAUDE.md hierarchies and custom slash commands
- Build MCP integrations connecting Claude to external tools and data sources
- Make daily production decisions about context management, cost optimization, and failure recovery
You should wait if you:
- Are still learning Claude API fundamentals
- Have primarily used Claude through the chat interface
- Have not yet shipped a production application using Claude
Roles that benefit most: AI Solution Architect, Multi-Agent System Engineer, Machine Learning Application Developer, Developer Enablement Engineer, and Technical Product Manager for AI integrations.
Explore the best AI certs in 2026: AI Security Certifications To Pursue In 2026
Is the CCA-F Worth It in 2026?
| Certification | Focus | Cost | Best For | Proctored? | Production Emphasis |
|---|---|---|---|---|---|
Anthropic CCA-F | Agentic architecture, MCP, Claude Code, reliability | $125 (free early for partners) | Claude-specific architects | Yes | Very High |
| AWS Certified AI Practitioner | Broad AWS AI services | ~$100 | Entry-level AWS AI roles | Yes | Medium-High |
| AWS Certified Machine Learning Engineer | SageMaker, advanced ML pipelines | ~$150 | Experienced AWS ML engineers | Yes | Medium-High |
| Google Professional ML Engineer | Google Cloud ML, Vertex AI | ~$200 | GCP-focused roles | Yes | High |
| Generic Prompt Engineering Certs | Basic prompting | Low/Free | Beginners | No | Low |
- As of mid-2026, the CCA-F is the only vendor-specific proctored credential focused on production Claude agentic systems. It provides strong enterprise signaling, especially at consulting firms running large Claude programs. Broader individual access is anticipated later in 2026.
Explore what it takes to earn AWS ML certification in the current year: Exploring Salary Trends and Career Worth of AWS Machine Learning Engineer Associate Certification
CCA-F Exam Format, Logistics, and Scoring
The exam includes 60 multiple-choice questions, each presenting one correct answer and three carefully engineered distractors. The distractors are not obviously wrong—they reflect real architectural mistakes that experienced engineers make.
Four of six predefined production scenarios are randomly assigned to each candidate. Each scenario provides roughly 15 questions tied to a shared narrative context. Because the scenario selection is random, candidates must prepare for all six.
Scoring uses a scaled system from 100 to 1,000. The passing threshold is 720. You do not need to pass each domain separately—your cumulative performance across all questions determines your final score. There is no penalty for guessing, so answer every question.
Logistics checklist:
- Block 2.5 hours in your calendar—you cannot pause once the exam starts
- ProctorFree requires a working webcam, microphone, and stable internet connection
- Maintain a clean desk with no second monitors
- Close all applications except the exam browser window
- Results, including a domain-by-domain breakdown, are delivered within two business days
Claude Certified Architect Exam Domains and Weighting
The CCA-F exam measures competency across five weighted domains. Understanding these weights is critical for prioritizing your study time.
| Domain | Weight |
|---|---|
| Agentic Architecture & Orchestration | 27% |
| Claude Code Configuration & Workflows | 20% |
| Prompt Engineering & Structured Output | 20% |
| Tool Design & MCP Integration | 18% |
| Context Management & Reliability | 15% |
Domain 1: Agentic Architecture & Orchestration (27%)
The largest and most challenging domain. It covers the full agentic loop (stop_reason: tool_use → execute → return → repeat), hub-and-spoke multi-agent coordination, task decomposition, and subagent context isolation. The core principle is that subagents do not automatically inherit any context from the coordinator. Every piece of information they need must be explicitly included in their task prompt.
Domain 2: Claude Code Configuration & Workflows (20%)
This domain tests practical mastery of Claude Code as a development environment. Key topics include the three-level CLAUDE.md hierarchy (user, project, directory), path-specific .claude/rules/ with YAML glob frontmatter, the -p flag for non-interactive CI/CD pipelines, and the context: fork skill option for isolated subagent execution.
Domain 3: Prompt Engineering & Structured Output (20%)
Covers few-shot prompting, validation-retry loops, the Message Batches API (50% cost savings, 24-hour window, no latency SLA), and multi-pass review architecture. A critical distinction the exam draws: tool_use with JSON schemas eliminates syntax errors but does not prevent semantic errors—fabricated values and field mismatches still require programmatic validation on top.
Domain 4: Tool Design & MCP Integration (18%)
Model Context Protocol is Anthropic’s open standard for connecting Claude to external tools and data. The exam tests how tool descriptions drive Claude’s tool selection, how to structure error responses with isRetryable and errorCategory fields, how to configure MCP servers in .mcp.json, and how to scope tools per agent role. The most common mistake: underestimating how much tool description quality influences Claude’s behavior.
Domain 5: Context Management & Reliability (15%)
Covers progressive summarization pitfalls, the lost-in-the-middle effect (key findings buried in long inputs get missed), transactional data preservation using persistent structured blocks, and deterministic escalation triggers. Combined, Domain 1 and agentic architecture and context management represent 42% of the exam.
The 6 Production Scenarios in the CCA-F Exam
Each candidate receives 4 of these 6 scenarios, selected randomly. Study all six.
| Scenario | Core Skills Tested |
|---|---|
| Customer Support Resolution Agent | Agent SDK, MCP, escalation patterns |
| Code Generation with Claude Code | CLAUDE.md, plan mode, slash commands |
| Multi-Agent Research System | Orchestration, subagents, context isolation |
| Developer Productivity Tools | Built-in tools, MCP servers, codebase navigation |
| Claude Code for CI/CD | Non-interactive -p flag, structured output |
| Structured Data Extraction | JSON schemas, tool_use, validation loops |
Each scenario provides a production narrative that grounds roughly 15 questions. The scenarios are not theoretical—they map directly onto the kinds of decisions architects face when building and maintaining Claude-powered systems at enterprise scale.
How to Prepare and Pass the Claude Certified Architect (CCA-F)
Anthropic’s preparation philosophy is straightforward: hands-on practice over passive reading. Every official resource is free.
Phase 1 — Foundations (Week 1)
Start with the agentic loop mental model. It underpins the majority of the exam. Make sure you can trace the full cycle: request → check stop_reason → execute tool → return result → repeat.
Courses:
- Claude 101 on Anthropic Academy — baseline knowledge the exam assumes
- Building Applications with the Claude API — the flagship 8+ hour course; this is non-negotiable
Phase 2 — Claude Code Hands-On (Week 2)
Install Claude Code and use it on a real project. Configure CLAUDE.md files from scratch, write custom slash commands, and run Claude Code in a CI/CD pipeline with the -p flag.
Course: Claude Code in Action on Anthropic Academy
Phase 3 — MCP & Tool Design (Week 3)
Build at least one MCP server yourself. Understand how tools are described, how errors are structured with isRetryable and errorCategory, and how to configure project-level vs. personal MCP servers.
Courses: Introduction to Model Context Protocol + MCP Advanced Topics on Anthropic Academy
Phase 4 — Prompt Engineering & Structured Output (Week 4)
Focus on few-shot prompting, validation-retry loops, and the Batch API’s tradeoffs. The exam tests which technique to apply in which production context—not just whether you know the techniques exist.
Phase 5 — Multi-Agent Systems & Context Management (Week 5)
This is where everything converges. Coordinator-subagent patterns, the Task tool, PostToolUse lifecycle hooks, context window optimization, and structured error propagation. This combined area is 42% of the exam.
Phase 6 — Practice Exams & Review (Week 6+)
The official Exam Guide (available as a PDF inside Skilljar after registration) contains 12 sample questions with detailed answer explanations—including explanations of why wrong answers are wrong. Study those carefully.
Then take the official 60-question practice exam. Aim for 900+ before scheduling your proctored session.
Free community resources:
- claudecertifications.com — free study guides, 25 practice questions, and a 12-week prep plan
- github.com/anthropics/courses — hands-on notebooks covering API fundamentals, prompt engineering, evaluations, and tool use
Registration, Cost, and the Anthropic Partner Network
The CCA-F exam is currently exclusive to Anthropic partner organizations and their employees. Individual registration without a qualifying organization is not yet available, though broader public access is widely anticipated later in 2026.
How to register:
- Apply to the Claude Partner Network at claude.com/partners (free; no minimum company size)
- Once approved, access Anthropic Academy at anthropic.skilljar.com
- Complete the recommended preparation courses
- Download the official CCA Foundations Exam Guide PDF
- Take the official 60-question practice exam
- Register and schedule your proctored exam slot
Cost:
- Free for the first 5,000 partner company employees (Early Access)
- $125 USD after Early Access closes
The exam is delivered through Anthropic’s certified partner network. Fractal Analytics is one of the partners currently offering access to the program.
Delve into the details to know the overall investment required for taking the AI certifications: AI Certification Cost In 2026: What To Expect
Benefits, Digital Badge, and Career Impact
Passing the CCA-F yields a shareable digital CCA-F badge verified by Anthropic. You can add it directly to your LinkedIn profile, resume, or professional portfolio. The credential validates applied expertise in agentic architecture, multi-agent orchestration, MCP integration, and production reliability—skills that are genuinely difficult to demonstrate without a structured assessment.
The credential carries enterprise weight for a specific reason: the proctored, closed-book format makes it resistant to the credential inflation that typically erodes certification value over time. The same structural dynamic made AWS certifications meaningful between 2013 and 2018.
At the enterprise level, Anthropic’s investment behind this certification is substantial. Accenture is training approximately 30,000 professionals on Claude. Cognizant has opened access to Claude for approximately 350,000 employees. Deloitte is embedded as an enterprise deployment partner. At that scale, the CCA-F is quickly becoming a baseline expectation for architects working on Claude implementations at major consulting firms.
CCA-F holders receive priority access to every new certification tier Anthropic launches through 2026 and beyond. The Foundations credential is the entry point to a growing stack.
Get insights on: Explore The Best AI Certifications For Beginners: A Great Start To Your AI Career In 2026
Common Pitfalls and Expert Tips for CCA-F
The exam’s wrong answers are not random. They are carefully constructed anti-patterns—exactly the mistakes experienced engineers make before they fully understand production implications. Knowing what NOT to do is genuinely half the exam.
Seven anti-patterns the exam tests repeatedly:
- Prompt-based enforcement for critical rules. Correct answer: use programmatic hooks and deterministic validation. Prompts can be ignored; code cannot.
- Self-reported confidence for escalation. LLM self-assessed confidence is poorly calibrated. Use deterministic threshold logic instead.
- Batch API for blocking workflows. The Batch API has no latency SLA. Any user-facing, real-time workflow must use the synchronous API.
- Larger context window to fix attention problems. Window size does not solve attention distribution. Split processing into focused, per-file passes.
- Silent failure on subagent error. Return structured error context—errorCategory, isRetryable, partial results—so the coordinator can recover.
- All tools available to all agents. Scope tools to each agent’s specific role. Excess tooling causes reasoning overload and degrades performance.
- Flat multi-agent topology. Use hub-and-spoke design with a central coordinator. Subagents never communicate directly with each other.
Expert test-taking tips:
- Flag difficult questions and return to them—time management matters at 2 minutes per question
- When a scenario shows tool misrouting, the correct first fix is almost always improving tool descriptions, not adding a routing classifier
- When a scenario shows extraction errors, distinguish whether data is present but mis-extracted (retry-with-feedback works) or absent from the source (retrying won’t fix it)
- Programmatic enforcement is almost always preferred over prompt instructions for anything financial, security-critical, or compliance-sensitive
What Are the Official Resources for CCA-F Exam Preparation?
All official preparation resources are free and available through Anthropic Academy at anthropic.skilljar.com. The platform launched March 2, 2026—ten days before the exam—with 13 self-paced courses.
Key courses mapped to exam domains:
| Course | Domain Coverage |
| Claude 101 | Foundational knowledge baseline |
| Building Applications with the Claude API (8+ hrs) | Agentic architecture, API integration, RAG |
| Claude Code in Action | Claude Code configuration and CI/CD |
| Introduction to Model Context Protocol | MCP fundamentals |
| MCP Advanced Topics | Building MCP servers and clients |
| Claude Code Sub-Agents | Context management and delegation |
The official CCA Foundations Exam Guide PDF and the official 60-question practice exam are both available inside Skilljar after registration. These are your highest-priority preparation materials.
Start Preparing for the CCA-F Now
The CCA-F represents something genuinely new: a rigorous, proctored credential that measures architectural judgment under production conditions—not just familiarity with Claude’s features.
For practitioners already building with Claude, the certification validates skills you have developed and sharpens understanding of the exact failure modes that matter in real deployments. For those earlier in the journey, the free Anthropic Academy courses are worth completing regardless of certification plans.
The first-mover window is real. Engineers who certify now establish credibility before this credential becomes table stakes. Begin by applying to the Claude Partner Network at claude.com/partners, then enroll at anthropic.skilljar.com to access all 13 free preparation courses.
Frequently Asked Questions
What is the Claude Certified Architect (CCA-F) certification?
The Claude Certified Architect – Foundations (CCA-F) is Anthropic’s first official technical certification, launched March 12, 2026. It is a 60-question, 120-minute, proctored exam that assesses candidates’ ability to design and deploy production-grade Claude applications using the Claude Agent SDK, Claude Code, the Claude API, and the Model Context Protocol (MCP). The passing score is 720 out of 1,000.
How much does the CCA-F exam cost?
The CCA-F exam is free for the first 5,000 employees of partner companies during Early Access. After Early Access closes, the exam fee is $125 USD.
Who can take the Claude Certified Architect exam?
As of June 2026, the CCA-F exam is exclusive to employees of organizations enrolled in the Claude Partner Network. Joining the Claude Partner Network is free and open to any organization actively bringing Claude to market. Broader individual access is expected later in 2026, but no official date has been confirmed.
What domains does the CCA-F exam cover?
The CCA-F exam covers five domains: Agentic Architecture & Orchestration (27%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%), Tool Design & MCP Integration (18%), and Context Management & Reliability (15%).
How many scenarios are on the CCA-F exam?
The official exam includes a pool of six production scenarios. Each candidate is randomly assigned four of the six scenarios, with approximately 15 questions tied to each. All six scenarios must be studied because the assignment is random.
What is the passing score for the Claude Certified Architect exam?
The passing score is 720 out of 1,000 on a scaled scoring system. Candidates do not need to pass each domain separately—cumulative performance across all 60 questions determines the final score.
How do I register for the CCA-F exam?
Apply to the Claude Partner Network at claude.com/partners, then access Anthropic Academy at anthropic.skilljar.com. Complete the recommended courses, download the Exam Guide PDF, take the official practice exam, and schedule your proctored session through the portal.
What happens after I pass the Claude Certified Architect exam?
Successful candidates receive a domain-by-domain score report within two business days and a shareable digital CCA-F badge verified by Anthropic. The credential is valid indefinitely until a superseding version is released. CCA-F holders also receive priority access to future certification tiers Anthropic plans to release through 2026.
Is there a free practice exam for the CCA-F?
Yes. Anthropic provides an official 60-question practice exam through the Skilljar portal. It uses the same scenario-based format as the real exam and includes post-answer explanations. Aim for a score of 900 or above on the practice exam before scheduling your actual session.
What preparation resources does Anthropic offer for the CCA-F?
All official preparation resources are free. Anthropic Academy at anthropic.skilljar.com offers 13 self-paced courses, including the flagship “Building Applications with the Claude API” course (8+ hours). The official CCA Foundations Exam Guide PDF is also available inside Skilljar after registration.
