Anthropic Skilljar: Academy Courses vs CCA-F Exam

Anthropic delivers its public learning catalog through Skilljar, the same hosted LMS platform used by many enterprise software vendors. If you have searched "anthropic skilljar" you have likely landed on academy.anthropic.com and wondered how those free courses connect to the Claude Certified Architect, Foundations (CCA-F) exam. The short answer: the Academy is a useful on-ramp, but it was not designed to the exam blueprint. This guide maps every public course to the five weighted domains, flags the gaps, and tells you exactly where to spend your remaining study hours.

What is the Anthropic Skilljar platform and who is it for?

Anthropic Academy runs on Skilljar and is free to access. You register with an email address; you do not need an existing Anthropic Console account to enrol. The catalog is aimed at a wide audience, from product managers exploring Claude 101 to engineers building with the API. Content is Anthropic-product-specific: you will not find general LLM theory or comparisons with competing models.

The courses are self-paced video modules with short knowledge checks. Completion certificates are issued by Skilljar on Anthropic's behalf. These certificates are distinct from the CCA-F credential, which is a proctored, scenario-based exam scored on a 100-to-1000 scale with a passing mark of 720.

How do the Academy courses map to the five CCA-F domains?

The CCA-F blueprint has five domains with distinct weightings. The table below shows how the public Academy courses align to each domain and flags coverage quality honestly.

Domain 1 carries the heaviest weight at 27%, and it is precisely where the Academy courses are thinnest. Topics such as hub-and-spoke architecture, subagent context isolation, and coordinator dynamic subagent selection require a level of scenario-based reasoning that the video modules do not drill.

Which Academy course should a solutions architect take first?

Start with Claude 101 regardless of your experience level. It establishes the mental model for the Messages API, token budgets, and model tiers that every other course assumes. From there, the recommended sequence for a solutions architect targeting the CCA-F is:

1. Claude 101 (foundational API and model concepts)
2. Prompt Engineering (structured output, few-shot patterns)
3. MCP (Model Context Protocol server setup and tool registration)
4. Claude Code 101 (CLI, CLAUDE.md, workflows)
5. Subagents (multi-agent orchestration patterns)

This sequence mirrors the exam domain weighting roughly in reverse difficulty: you build vocabulary before tackling the architecturally complex Domain 1 material. The MCP course is worth completing before Claude Code 101 because Claude Code uses MCP servers internally, and understanding MCP scoping hierarchy makes the configuration lessons far less confusing.

Are the Anthropic Academy courses truly free?

Yes. Every course on the Anthropic Skilljar platform is free at the time of writing. You create a Skilljar account with an email address and enrol immediately. There are no paid tiers within the Academy catalog itself.

The CCA-F exam, by contrast, costs $99 per attempt. Tiered Anthropic partners receive a discounted first attempt through the Claude Partner Network. As of 3 June 2026, more than 40,000 firms have applied for partner status and more than 10,000 individuals hold the certification.

How much MCP depth does the exam actually require?

More than the Academy course alone provides. The CCA-F exam tests MCP at the level of production integration decisions, not just "how do I start a server." Specifically, the exam expects you to reason about:

• When to build a custom MCP server versus consuming an existing one (the build vs use decision)
• How isError flags propagate through multi-agent pipelines
• Environment variable expansion in MCP config files
• Tool description quality as the primary routing mechanism

The Academy MCP course covers server setup and basic tool registration well. It does not cover structured error metadata or the nuances of tool splitting for specificity that appear in exam scenarios.

A representative exam-style question looks like this:

Scenario: A coordinator agent calls a search tool and receives a response
with isError: true and an empty content array. The agent retries twice
with identical parameters and receives the same response.

Which action best reflects the exam's preferred approach?
A. Continue retrying with exponential back-off.
B. Escalate to a human operator with structured error context.
C. Silently skip the search step and proceed with available data.
D. Switch to a different model tier for the retry.

The exam consistently rewards deterministic, proportionate responses and root-cause tracing. Option B is correct here: two identical failures signal a systemic issue, not a transient one, and silent suppression (option C) is an explicit anti-pattern in the exam's scoring logic.

What do architects need to know about Claude Code for the exam?

The Claude Code domain (Domain 3, 20% weight) tests configuration hierarchy, workflow design, and the interaction between CLAUDE.md files and MCP server definitions. The Academy's Claude Code 101 course covers the CLI flags and basic slash commands adequately. The exam goes further.

Key concepts the course under-covers:

• The three-level configuration hierarchy (global, project-root, sub-directory) and how conflicts resolve
• Version control implications of committing CLAUDE.md versus keeping it in .gitignore
• Using --resume versus --fork versus a fresh session for different task types
• Hook pipelines for data normalisation before tool results reach the model

# Example: launching Claude Code with a specific MCP server and non-interactive mode
claude --mcp-config ./mcp.json -p "Analyse the failing tests in /src and output a JSON report" \
  --output-format json

The -p flag for non-interactive mode and --output-format json are both testable. Architects who have only used the interactive REPL in the Academy course often miss these in scenario questions.

What are the most common exam gotchas in the Academy-covered topics?

Even in areas the Academy covers well (Domains 4 and 5), several concepts trip up candidates who relied solely on the video modules.

The embeddings point catches a surprising number of candidates. The Academy's Claude 101 course focuses on the Messages API and does not explicitly state that Claude is not an embeddings model. The exam does test this distinction.

How does AI Skill Certs complement the Anthropic Skilljar catalog?

AI Skill Certs is an independent adaptive prep platform for the CCA-F exam. We are not affiliated with, endorsed by, or approved by Anthropic. What we do is map the gaps the Academy leaves.

Our concept library at /concepts covers 174 atomic concepts mapped to all five exam domains and all 30 task statements. Each concept is tagged to the domain weight, so you can see at a glance that Domain 1's agentic architecture material deserves proportionally more of your study time than Domain 5.

The adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold. That means the platform keeps serving you Domain 1 scenarios until your probability of mastery crosses 90%, rather than moving on after a single correct answer. Practice exams are 60 questions, scored 100 to 1000 with 720 as the passing bar, identical to the real exam format.

Our suggested study path for someone who has completed the full Anthropic Academy catalog:

1. Take a diagnostic practice exam to identify your weakest domains.
2. Work through the concept library for any domain scoring below 600 on the diagnostic.
3. Focus additional time on tool design and MCP integration and agentic architecture, which together account for 45% of the exam.
4. Re-take the practice exam until you consistently score above 750 (a 30-point buffer above the pass mark).
5. Book the proctored exam when two consecutive practice attempts exceed 750.

The Academy courses are a genuine asset for building vocabulary and intuition. They are not sufficient on their own for the 720 pass mark, particularly for candidates without prior multi-agent systems experience.