Claude Certified Developer Foundations: 2026 Exam Guide

The Claude Certified Developer Foundations (CCDV-F) exam is Anthropic's developer-track certification, one of four proctored exams in the Claude Partner Network programme launched 12 March 2026. At $125 per attempt, it tests whether candidates can actually build with Claude: integrate the API, design tools, manage context, select the right model, and handle security in production. It is not a syntax quiz.

In this guide we cover the complete eight-domain blueprint, the study priorities the weights signal, and how to prepare when official practice material remains limited.

What is the CCDV-F exam?

CCDV-F is a $125 exam delivered by Pearson VUE, either online-proctored or at a test centre. Scored on a 100 to 1000 scale with 720 as the passing score, it assesses practical developer skills across eight domains.

Unlike the architect track (CCAR-F), CCDV-F has no scenario bank. Items are written directly against the skills in each domain, so every exam draws from the full eight-domain blueprint. There is no tactical shortcut of preparing for a subset.

Per the official CCDV-F exam guide, every item is scenario-based and tests practical judgment rather than factual recall. That shapes how you should prepare: pattern recognition and trade-off reasoning matter more than memorising API syntax.

What does the exam format look like?

53 items in 120 minutes. Items are multiple-choice and multiple-response; each item tells you how many responses to select. Your score report gives pass or fail, the scaled score, and percent-correct by domain. Anthropic does not publish the raw-to-scaled conversion, so no external source can state exactly how many questions constitute a passing raw score.

Tiered Claude Partner Network partners receive discounted first attempts. If your employer is a partner, confirm eligibility before paying the standard rate.

Which domains are tested, and how much does each count?

The eight-domain breakdown tells you precisely where to invest study time. Applications and Integration at 33.1% is the dominant domain. By comparison, Claude Code (3.1%) and Eval, Testing, and Debugging (2.6%) together represent less than 6%.

Domains 2, 5, and 1 together account for roughly 65% of the exam. Domains 6 and 8 add another 21.6%. A proportionate study plan focuses primarily on those five domains while ensuring you can still score on Domains 3, 4, and 7.

How should I study Domain 2: Applications and Integration?

Domain 2 is the exam's centre of gravity. At 33.1%, it shapes more of your score than any other domain. It tests real API integration work: structuring requests correctly, parsing responses, handling errors, managing authentication, and building pipelines that hold up under production load.

Three areas appear consistently in Domain 2 scenarios:

1. Request and response structure. Understanding the complete lifecycle of an API call, including how stop_reason field inspection feeds into downstream application logic, is foundational.

2. Streaming and asynchronous patterns. Production applications rarely block on a single synchronous call. Candidates should understand when to stream, how to handle partial outputs, and how to maintain state across multi-turn conversations.

3. Error handling and retries. The exam rewards candidates who distinguish retryable rate-limit errors from hard failures and who design applications that degrade gracefully rather than crash silently.

A minimal working API call illustrates the baseline knowledge this domain tests:

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain the difference between max_tokens and token budgets."}
    ]
)

print(response.stop_reason)   # "end_turn", "max_tokens", "tool_use", etc.
print(response.content[0].text)

The exam does not ask you to write code from scratch. It asks whether you can diagnose why an integration behaves unexpectedly and select the right fix. That requires hands-on SDK experience alongside concept-level understanding.

What about Domains 1 and 5, the other major weights?

Domain 1 (Agents and Workflows, 14.7%) tests agentic system design: when to use single-agent patterns versus parallel subagent spawning, how to handle tool results in a loop, and how to avoid failure modes that make systems unreliable at scale.

Hub-and-spoke architecture and multi-agent error routing are representative concept areas. Exam items in this domain frequently describe a broken agentic loop and ask candidates to identify the root cause and select a proportionate fix.

Domain 5 (Model Selection and Optimisation, 16.8%) tests cost and latency trade-offs. Anthropic's model family spans a wide range of capability and price. A typical Domain 5 scenario presents a production constraint, such as a high-volume pipeline with a tight latency budget, and asks which combination of model, caching, and batching approach is most appropriate.

What does Domain 8 (Tools and MCPs) test?

Domain 8 accounts for 10.6% of the exam and covers the Model Context Protocol and tool design. Items test whether candidates can write tool descriptions that Claude selects correctly, handle tool errors without breaking the application loop, and scope tools appropriately for multi-agent systems.

Writing effective tool descriptions is foundational. A poorly described tool leads to misrouting, and many Domain 8 scenarios present exactly that failure mode. The fix is almost always in the description rather than the code.

A well-structured tool definition illustrates what the exam expects:

{
  "name": "get_order_status",
  "description": "Returns the current fulfilment status and estimated delivery date for a specific order. Use when the user asks about an existing order. Do not use for product availability or pricing questions.",
  "input_schema": {
    "type": "object",
    "properties": {
      "order_id": {
        "type": "string",
        "description": "The unique order identifier, format ORD-XXXXXXXX"
      }
    },
    "required": ["order_id"]
  }
}

The description field does the selection work. It tells Claude when to call the tool and, just as importantly, when not to. Both parts are necessary.

Should I study the smaller domains?

Domains 3 and 4 are small at 5.7% combined, but ignoring them is a mistake if you are sitting close to the 720 pass mark. A few targeted items can shift a marginal result.

For Domain 3 (Claude Code, 3.1%), focus on how Claude Code operates as a developer tool, how to configure it for a project, and how it integrates into a team workflow. Claude Code Configuration and Workflows covers the key configuration concepts.

For Domain 4 (Eval, Testing, and Debugging, 2.6%), the exam tests evaluation design: how to construct test cases that catch real failures, how to interpret evaluation output, and how to debug unexpected model behaviour. These skills appear in every production deployment.

Domain 6 (Prompt and Context Engineering, 11.0%) is larger than Domains 3 and 4 combined. It covers prompt design and context management techniques that make applications reliable. Strong candidates understand not just how to write prompts but why certain structures produce more consistent outputs. Prompt Engineering and Structured Output is the directly relevant concept area.

Domain 7 (Security and Safety, 8.1%) deserves attention beyond its weight. Agentic applications that take real-world actions carry real risk. The exam tests whether candidates understand prompt injection, data leakage vectors, and how to design systems that fail safely rather than silently.

What prep options exist for CCDV-F?

The official exam guide from Anthropic describes domains and task statements but does not include a large question bank. Public sample material remains limited as of the 12 March 2026 launch date.

Our CCDV-F prep is live on AI Skill Certs. Adaptive study, Archie tutoring, and practice exams for the developer track are available today. Practice exams mirror the real format: 53 questions, scored 100 to 1000, with 720 as the passing bar. Archie, our Socratic tutor, uses graduated hints rather than direct answers to build the pattern-recognition that scenario-based items require.

As of 3 June 2026, the Claude Partner Network has 10,000+ certified individuals across all four tracks, with 40,000+ partner applicant firms. The CCDV-F concept library at /concepts is not yet live, but adaptive study and practice exams are available now.

AI Skill Certs is an independent platform and is not affiliated with, endorsed by, or approved by Anthropic.

How does CCDV-F compare with CCAR-F?

Both exams cost $125 and require 720 to pass. The differences reflect genuinely different job profiles.

CCDV-F is the right starting point for developers who work directly with the API and ship Claude-powered features. CCAR-F suits architects designing multi-agent orchestration layers. Both credentials are part of a $100 million Partner Network programme and complement rather than compete with each other.