Anthropic Academy Courses: CCA-F Exam Prep Guide 2026

Anthropic Academy courses are free, self-paced, and certificate-bearing, making them the most obvious starting point for anyone preparing for the Claude Certified Architect, Foundations (CCA-F) exam. But "obvious starting point" is not the same as "sufficient preparation." This guide maps every major Academy track to the five CCA-F domains, flags the gaps, and tells you where to spend your remaining study hours.

What are Anthropic Academy courses, and are they really free?

Yes, the courses are genuinely free. You register through Anthropic's Skilljar-hosted Academy portal using either an existing Anthropic account or a new Skilljar account; no credit card is required. Completion certificates are issued per course, though these are distinct from the CCA-F credential itself.

The catalogue has grown since the Academy launched. Early coverage described 13 courses; the current catalogue lists 17 tracks spanning introductory, developer, and partner-facing content. Anthropic updates the Academy without versioned release notes, so the precise count can shift. Check Anthropic's Academy portal directly for the live list rather than relying on any cached screenshot.

How does the Academy catalogue map to the five CCA-F domains?

The CCA-F exam covers five weighted domains. The table below shows how the main Academy tracks align, along with each domain's exam weight.

Domain 1 carries the heaviest weight at 27%, yet the Academy's agentic content is spread across multiple courses rather than consolidated. Candidates who treat Claude Code 101 as their sole agentic preparation will miss orchestration concepts such as hub-and-spoke architecture and coordinator dynamic subagent selection that appear in scenario questions.

Which course should a solutions architect take first?

Start with Claude 101 regardless of your prior experience. It establishes the mental model for how Claude processes context, handles ambiguity, and applies safety constraints. Skipping it to jump straight to Claude Code 101 leaves gaps in Domain 4 (Prompt Engineering & Structured Output, 20%) and Domain 5 (Context Management & Reliability, 15%).

After Claude 101, the recommended sequence for CCA-F candidates is:

1. Claude Code 101 (Domains 1 and 3)
2. MCP 101 (Domain 2)
3. Prompt Engineering for Developers (Domain 4)
4. AI Limitations, Safety & Security (Domain 5)
5. Building with Claude Agents (Domain 1 reinforcement)

This order mirrors the exam's descending domain weight. You front-load the highest-weighted material while the foundational concepts from Claude 101 are still fresh.

What does the MCP track cover, and is it enough for exam questions?

The MCP 101 course covers the Model Context Protocol specification, tool discovery, the MCP inspector, and basic integration patterns for connecting Claude to external systems. For Domain 2 (Tool Design & MCP Integration, 18%), this is a solid foundation.

However, the Academy course does not go deep on failure modes. Exam scenarios regularly test whether candidates can distinguish an access failure from a valid empty result, handle the isError flag pattern, or design tool descriptions that prevent misrouting. These are operational concerns that the introductory MCP track treats lightly.

Supplement the Academy MCP course with hands-on work: build a minimal MCP server, intentionally break tool discovery, and trace the error propagation. The exam rewards candidates who have reasoned through failure paths, not just those who can recite the happy-path flow.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/tmp/sandbox"],
      "env": {
        "READ_ONLY": "true"
      }
    }
  }
}

The snippet above shows a minimal MCP server configuration with an environment variable constraint. Exam questions on environment variable expansion in MCP config expect you to reason about what happens when READ_ONLY is absent or overridden at a different scope.

How deep does the Claude Code track go on subagents and reusable workflows?

Claude Code 101 introduces the three-level configuration hierarchy, CLAUDE.md files, and basic slash commands. The advanced Claude Code course extends this to subagent delegation, agent skills, and reusable workflow patterns.

For Domain 3 (Claude Code Configuration & Workflows, 20%), you need to understand:

• How configuration scopes interact (user, project, and sub-directory levels)
• When a skill defined in a CLAUDE.md file takes precedence over a system prompt instruction
• How to design reusable workflows that survive session resets

The Academy covers the first two points adequately. The third, session resilience, is where candidates most often lose marks. Summary injection for fresh sessions and the choice between resuming, forking, or starting fresh are Domain 1 and Domain 3 crossover topics that the Academy treats as separate concerns. The exam treats them as one integrated decision.

What is the difference between Claude Cowork and Claude Code?

Claude Cowork is Anthropic's collaborative workspace product, designed for teams working together in a shared interface. Claude Code is the agentic coding tool that runs in your terminal, reads and writes files, executes commands, and integrates with your local development environment.

For the CCA-F exam, Claude Code is the relevant product. Exam scenarios assume a local or CI/CD context where the candidate is configuring Claude Code's behaviour through CLAUDE.md files, hooks, and MCP server definitions. Claude Cowork does not appear in the current exam guide.

The distinction matters because some Academy marketing materials group both under the "Claude for teams" umbrella. If you find yourself reading about shared workspaces and real-time collaboration, you have drifted into Cowork territory and should redirect to the Claude Code tracks.

What AI limitations, safety, and security topics appear on the exam?

Domain 5 (Context Management & Reliability, 15%) and Domain 1 both surface safety and reliability concerns. The Academy's AI Limitations, Safety & Security course covers:

• Hallucination and confidence calibration
• Trusted versus untrusted input sources
• Custom skill boundaries and permission scoping
• Prompt injection vectors in agentic pipelines

The exam applies these principles in scenario form. A typical question presents an agentic pipeline where a subagent receives instructions from an external tool result and asks whether the pipeline should execute, pause for human review, or reject the instruction. The correct answer almost always traces to whether the instruction source is trusted and whether the action is reversible. The Academy course gives you the vocabulary; the high-stakes enforcement decision rule concept gives you the decision framework.

Do AWS Bedrock and Google Cloud Vertex AI tracks matter for CCA-F?

Anthropic has published Academy content covering Claude on AWS Bedrock and Google Cloud Vertex AI. These tracks are relevant for practitioners deploying Claude in cloud-managed environments, but they are not tested on the CCA-F exam as of the exam guide published at launch on 12 March 2026.

The CCA-F exam tests Claude's native APIs, the Messages API, tool use, MCP, and Claude Code, rather than cloud-provider-specific SDK wrappers. If your role involves Bedrock or Vertex AI deployments, the cloud tracks are professionally useful. For exam preparation specifically, they are optional reading.

How should you fill the gaps the Academy leaves?

The Academy is a strong foundation but it was not designed as a CCA-F study programme. Three categories of gap appear consistently:

Scenario reasoning. The exam's 60 scenario-based questions require you to choose between plausible options, not just recall definitions. The Academy's courses are largely expository. Supplement with timed practice under exam conditions.

Concept precision. Terms like parallel subagent spawning, structured context passing, and agentic loop anti-patterns appear in exam distractors with subtle variations. The Academy introduces these ideas but rarely tests the boundary cases.

Domain 1 depth. At 27% of the exam, Agentic Architecture & Orchestration is the single most important domain. The Academy's agentic content is spread across at least three courses and does not present a unified orchestration model. Our Agentic Architecture & Orchestration concept library maps all 30 task statements to atomic concepts, which is a more efficient way to close this gap than rewatching course videos.

The CCA-F passing score is 720 on a 100-to-1000 scale, per Anthropic's exam guide. That is a meaningful bar. Candidates who complete the Academy courses and then move directly to the exam without scenario practice consistently report that the questions felt harder than the course material suggested. Build in at least two full 60-question practice sessions, scored on the same 100-to-1000 scale, before booking your attempt.

AI Skill Certs is an independent prep platform; we are not affiliated with or endorsed by Anthropic. Our concept library at /concepts covers 174 atomic concepts mapped to all five domains and 30 task statements, and our adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold to surface the concepts where you are weakest. If you have finished the Academy tracks and want to know where your gaps actually are, that is where we would point you next.