Claude 101 Course: Developer Study Path for CCDV-F

The CCDV-F (Claude Certified Developer, Foundations) is the developer track of Anthropic's certification programme, and a structured claude 101 course built around its eight domain weights gives developers the most direct path to the 720 passing score. The exam costs $125 per attempt, covers 53 items in 120 minutes, and launched 12 March 2026 as part of the Claude Partner Network's four live certification tracks.

This guide maps every domain, explains what each one tests in production terms, and gives you a weighted study allocation so you know where to put your time first.

What is the CCDV-F exam format?

The Claude Certified Developer, Foundations exam carries the official code CCDV-F. It is one of four tracks in the Claude Partner Network alongside the Architect Foundations (CCAR-F, $125), Associate Foundations (CCAO-F, $99), and Architect Professional (CCAR-P, $175) exams.

A key structural difference from the Architect exam: CCDV-F has no scenario bank. Per Anthropic's official CCDV-F exam guide (2026-07-08), items are written directly against the skills in each domain rather than drawn from a rotating pool of scenarios. Every question is production-oriented. You will be asked to choose the right tool, model, or architecture for a realistic build decision, not to recall a definition.

The credential is valid for 12 months from the award date. As of 3 June 2026, more than 10,000 individuals hold a Claude certification across the four tracks, and the Claude Partner Network has received applications from more than 40,000 firms.

What are the eight CCDV-F domains and their weights?

The exam spans eight domains with exact weights published in the exam guide:

Domain 2 alone accounts for roughly one in three questions on the exam. Any claude 101 course that does not anchor its first weeks there is misallocating study time.

What does a claude 101 course cover for Applications and Integration (33.1%)?

Domain 2 is the weight-bearing wall of the CCDV-F exam. It covers the full lifecycle of a Claude-backed application: Messages API request construction, response parsing, streaming, batch processing, schema design, and configuration management.

A question in this domain might present a Python client making a Messages API call and ask which change reduces latency for a high-throughput use case, or show a JSON schema and ask whether it prevents a known hallucination pattern under edge-case inputs. The exam rewards builders who have moved past "it works in development" to "this holds in production at scale."

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Extract the invoice number and total as JSON."}
    ]
)
print(response.content)

The Prompt Engineering & Structured Output concepts are directly relevant here, covering schema-driven extraction, output reliability, and prompt constraints that appear across Domain 2 scenarios.

Batch versus synchronous routing is a recurring scenario type. The Batch API suits workloads where latency is not a constraint and cost per token matters. Synchronous streaming suits user-facing applications where time to first token drives experience. The exam expects you to reason from those constraints to a confident recommendation without hedging.

How does Model Selection and Optimization (16.8%) fit the course?

At 16.8%, Domain 5 is the second-heaviest domain. The exam tests whether you can match a given workload to the right model tier. The decision variables are latency requirements, context length, reasoning depth, and cost per token.

A support-ticket classifier that runs 10,000 times per day calls for a different model than a weekly contract summarisation job. The exam expects you to reason from those constraints to a model recommendation, not to memorise benchmark tables.

Cost optimisation is a connected skill. Prompt caching, batch routing, and token-trimming all appear under Domain 5. Candidates who can articulate the tradeoff between synchronous and batch API in both cost and latency terms are well prepared for this domain.

A typical scenario: a developer needs sub-500ms responses for an autocomplete feature. Which model and which API mode delivers that with acceptable quality? Preparation requires understanding the characteristics of each model tier and mapping those to scenario requirements. Together, Domains 2 and 5 represent 49.9% of the exam, making them the right starting point for any structured study plan.

What do Agents and Workflows questions test (14.7%)?

Domain 1 at 14.7% tests multi-step orchestration. The core skills are building an agentic loop that terminates correctly, routing between subagents, handling tool errors without dropping context, and choosing between a fixed sequential pipeline and a dynamic adaptive one.

The patterns covered in Agentic Architecture & Orchestration are relevant preparation here. Loop termination is worth extra attention: a significant share of agentic loop bugs trace to ignoring the stop_reason field and either looping forever or halting prematurely.

A common scenario: a coordinator agent spawns three subagents in parallel to gather data, then synthesises results. One subagent returns an error. What should the coordinator do? The exam rewards candidates who choose "retry once, then escalate with structured context" over those who swallow the error or immediately abort the task.

{
  "type": "tool_result",
  "tool_use_id": "toolu_01A2B3C4D5E6F7",
  "content": "Error: upstream API returned 503",
  "is_error": true
}

That is_error: true flag is the signal the exam will ask you to act on. The downstream handling choice is the testable judgment call.

What does a claude 101 course need from Prompt and Context Engineering (11.0%)?

Domain 6 at 11.0% covers prompt technique selection alongside context management. The Context Management & Reliability patterns are testable here: few-shot example construction, system prompt structure, context window management, and the decision between prompt-based and programmatic enforcement.

A representative scenario: a developer has a 200,000-token context window and a task that could fill it. When does filling the window help, and when does it dilute the model's attention on the critical section? The exam rewards candidates who know that longer is not always better and can explain why.

Context compaction strategies, summary injection, and the stale context problem all appear in this domain. If you can explain why a long-running agentic session might degrade in quality even with a large context window, you understand Domain 6.

How much do Tools and MCPs matter for the exam (10.6%)?

Domain 8 at 10.6% covers tool description writing, MCP server integration, the tool_choice configuration parameter, and the distinction between MCP resources, tools, and prompts.

The Tool Design & MCP Integration patterns cover the highest-leverage skill here: tool description quality. A poorly written description causes tool misrouting, and the exam will ask you to diagnose why a model reached for the wrong tool and how to fix it with the smallest possible change.

MCP scoping is also testable. Which configuration scope belongs at the project level versus the user level? When should you build a custom MCP server rather than adopt an existing one? These are judgment questions that require practical familiarity with MCP integration patterns, not just awareness that MCP exists.

How should a claude 101 course approach Security and Safety (8.1%)?

Domain 7 at 8.1% tests prompt injection awareness, output validation, safe defaults, and when to escalate to human review. The exam consistently rewards candidates who choose the conservative, deterministic option when safety is a concern.

If a question presents a scenario where a model could act autonomously on irreversible actions without a confirmation step, the correct answer is almost always to add the gate. Proportionate fixes, root-cause tracing, and human-in-the-loop checkpoints score better than clever workarounds. This principle generalises across the whole CCDV-F exam: when stakes are high, choose deterministic enforcement.

Recent practitioner interest in model misuse and live-system testing has made this domain feel increasingly exam-relevant. The exam is not asking you to exploit vulnerabilities; it asks whether your system design resists them by default.

What does the exam test for Claude Code (3.1%) and Evals (2.6%)?

Domain 3 (Claude Code) at 3.1% and Domain 4 (Eval, Testing, and Debugging) at 2.6% are the lightest domains by weight, but neither is safely skipped.

The Claude Code Configuration & Workflows coverage maps to Domain 3. The three-level configuration hierarchy (user, project, and session settings), CLAUDE.md instruction files, and the version control implications of project-level config are the testable areas. The exam will ask which configuration scope is correct for a given team setup, not ask you to recite keyboard shortcuts.

For Domain 4, the exam tests whether you can design a minimal eval that catches a known failure mode, interpret a score report to identify the weakest domain, and distinguish between a systematic model error and a one-off edge case. The weight is small, but the skill threads through every other domain: building trust in your Claude application through structured testing is what separates production-grade developers from those who ship on instinct.

How should you allocate study time across the course?

Follow the domain weights closely. The CCDV-F exam distributes difficulty in direct proportion to those percentages:

Start with Domain 2 (Applications and Integration) and Domain 5 (Model Selection and Optimization). Once those are solid, work through the remaining domains in descending weight order.

AI Skill Certs is an independent adaptive prep platform for the CCDV-F exam, not affiliated with or endorsed by Anthropic. The platform provides adaptive study sessions, Archie tutoring (a Socratic guide that uses graduated hints rather than direct answers), and a 53-question practice exam scored on the same 100-to-1000 scale with 720 as the pass mark. The adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold, so the system keeps presenting Domain 2 material until you demonstrate consistent accuracy before shifting focus to weaker areas.

The CCDV-F credential is valid for 12 months from the award date. Sitting the exam within four to six weeks of completing a structured course gives the best chance of that window staying useful before the material drifts from current.