AI Engineer Certification: The CCDV-F Complete Guide

If you are searching for an ai engineer certification that tests production-grade skills rather than trivia recall, the Claude Certified Developer, Foundations exam (CCDV-F) is the most direct option in Anthropic's credential stack. Launched 12 March 2026 as part of the Claude Partner Network, it costs $125 per attempt, runs 53 items in 120 minutes, and requires a scaled score of 720 out of 1000 to pass. This guide covers every domain, the study approach that matches the exam's practical focus, and the platform resources we have built to support you.

What is the CCDV-F exam, and how does it differ from the Architect track?

The CCDV-F is the developer track of Anthropic's Foundations certification programme. Where the Claude Certified Architect exam (CCAR-F) tests system design and orchestration judgment across five domains, CCDV-F is narrower in scope but deeper in implementation detail. It has 53 items rather than 60, and it does not draw from a scenario bank; every item is written directly against the skills listed in each domain. Both exams share the same 120-minute time limit, the same 100-to-1000 scoring scale, and the same 720 passing threshold.

The credential is valid for 12 months from the date it is awarded. There are no published prerequisites; Anthropic does not require you to hold the Associate credential (CCAO-F) before sitting CCDV-F. If you have hands-on experience building Claude-powered applications, you can register directly.

What are the eight CCDV-F domains and how are they weighted?

The exam covers eight domains with exact fractional weights published in the official exam guide. Understanding the weighting is the single most efficient way to allocate study time.

Domain 2 (Applications and Integration) carries 33.1% of the exam on its own. That is more than three times the weight of Domain 3 (Claude Code) and Domain 4 (Eval, Testing, and Debugging) combined. Any study plan that treats all eight domains equally will leave marks on the table.

Domains 5 (16.8%) and 1 (14.7%) round out the top three. Together, Domains 1, 2, and 5 account for roughly 64.6% of the exam. We recommend spending at least two-thirds of your preparation time on those three areas before turning to the remaining five.

What does Domain 2 (Applications and Integration) actually test?

Domain 2 is the exam's centre of gravity at 33.1%. It tests your ability to build real Claude-powered applications: structuring API calls correctly, handling streaming responses, managing authentication, integrating Claude into existing software stacks, and designing reliable pipelines that behave predictably in production.

Expect items that present a partially written integration and ask you to identify the fault, choose the correct API parameter, or select the response-handling pattern that avoids a specific failure mode. The Messages API request-response cycle is foundational here; if you are not fluent in how a request is constructed and how the response object is parsed, this domain will cost you points.

A typical integration item might look like this:

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Summarise this document in three sentences."}
    ]
)

print(message.content[0].text)

Items will probe whether you understand what message.content contains, when stop_reason signals a tool call versus natural completion, and how to handle errors gracefully rather than letting exceptions propagate silently.

How deep does Domain 1 (Agents and Workflows) go?

Domain 1 (14.7%) tests workflow-versus-agent decisions, agentic loop design, tool use, and multi-agent coordination. The exam distinguishes between scenarios where a fixed sequential pipeline is the right choice and scenarios that genuinely require dynamic, model-driven decomposition.

Key concepts include agentic loop anti-patterns, stop_reason field inspection, and how to route errors across agent boundaries. Items frequently present a broken agent loop and ask you to diagnose the root cause rather than simply identify that something is wrong.

The exam rewards deterministic solutions over probabilistic ones when stakes are high. If an item describes a financial transaction workflow, the correct answer almost always involves a hard gate or a structured handoff rather than asking the model to decide whether to proceed.

{
  "type": "tool_use",
  "id": "toolu_01XFDUDYJgAACTvykiHJQSCj",
  "name": "get_account_balance",
  "input": {
    "account_id": "ACC-9921"
  }
}

When the model returns a tool_use block like the one above, your application must append the tool result correctly before the next turn. Tool result appending is a concrete skill the exam tests; getting the message structure wrong is a common source of loop failures.

What do you need to know about Model Selection and Optimisation (Domain 5)?

Domain 5 carries 16.8% of the exam and focuses on choosing the right Claude model for a given task, optimising token usage, managing latency, and controlling cost. Items present production scenarios with specific constraints (latency budget, cost ceiling, quality floor) and ask you to select the model or configuration that satisfies all three.

You will need to understand the trade-offs between Haiku, Sonnet, and Opus class models in terms of speed, capability, and cost, and to apply those trade-offs to realistic workloads. Prompt caching, context window sizing, and batching decisions all appear here.

How much Claude Code knowledge does CCDV-F require?

Domain 3 (Claude Code) carries only 3.1% of the exam weight, the lowest of all eight domains. You do not need deep Claude Code configuration expertise to pass CCDV-F. A working familiarity with what Claude Code is, how it integrates into a developer workflow, and the basics of the three-level configuration hierarchy is sufficient for the items in this domain.

This is a meaningful contrast with the CCAR-F Architect exam, where Claude Code configuration and workflows form an entire 20% domain. If you are deciding between the two tracks, the developer exam is the right choice if your day-to-day work is building integrations and applications rather than designing multi-agent systems.

How security-heavy is the exam?

Domain 7 (Security and Safety) carries 8.1%. That is not trivial, but it is also not the exam's primary focus. You need a solid understanding of prompt injection risks, safe integration patterns, and how to design systems that do not expose sensitive data through Claude's outputs. Items tend to be scenario-based: a described integration has a security flaw, and you must identify it and select the correct remediation.

The exam does not require you to memorise Anthropic's Acceptable Use Policy verbatim. It does require you to apply safety principles to realistic production scenarios, which is a different skill. Understanding when to add a guardrail versus when a guardrail is unnecessary overhead is the kind of judgment the items test.

How much MCP knowledge is tested in Domain 8?

Domain 8 (Tools and MCPs) carries 10.6%. Items cover tool definition, tool selection by the model, error handling in tool responses, and MCP integration patterns. The MCP isError flag pattern and structured error metadata are concrete skills that appear in this domain.

A common item type presents a tool definition with a vague or ambiguous description and asks you to identify why the model is selecting the wrong tool. The fix is almost always a description rewrite rather than a structural change to the tool itself, which reflects the low-effort, high-leverage principle that runs through the exam's design philosophy.

{
  "name": "search_documents",
  "description": "Search for documents matching a query. Use this tool when the user wants to find specific files, records, or content by keyword or phrase. Do not use this tool for listing all documents in a folder.",
  "input_schema": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "The search term or phrase to look for"
      }
    },
    "required": ["query"]
  }
}

The description above is deliberately specific about when to use the tool and when not to. That specificity is what the exam rewards.

What does an efficient CCDV-F study plan look like?

Given the domain weights, a proportionate study plan allocates time roughly as follows:

Start with Domain 2 because it is both the heaviest domain and the one most likely to surface gaps in your API knowledge. Build a small integration from scratch, break it deliberately, and fix it. That cycle of build-break-fix is the closest analogue to what the exam items test.

For Domains 1 and 8, work through the tool design and MCP integration concepts systematically. The exam's tool-related items are among the most predictable: they almost always hinge on description quality, error propagation, or the distinction between an access failure and a valid empty result.

Practice exams are the most reliable signal of readiness. Our platform's CCDV-F practice exams mirror the real format exactly: 53 items, scored 100 to 1000, with 720 as the passing bar. The adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold, so it surfaces your weakest concepts rather than cycling through material you already know. Archie, our Socratic tutor, guides you through items you miss with graduated hints rather than giving you the answer directly.

Is there a prerequisite, and do you need the Associate cert first?

No published prerequisite exists. Anthropic does not require you to hold the Claude Certified Associate, Foundations credential (CCAO-F, $99) before registering for CCDV-F. The two exams are separate tracks, not a ladder. If you have practical experience building Claude-powered applications, you can register for CCDV-F directly.

That said, if you are new to the Claude API entirely, the Associate exam covers foundational concepts that will appear implicitly throughout CCDV-F. Sitting CCAO-F first is a reasonable choice for developers who are still building their baseline fluency, but it is not required.

How is the exam delivered, and what happens if you do not pass?

CCDV-F is delivered online-proctored or at a Pearson VUE test centre. The $125 fee applies per attempt. Anthropic does not publish a free-retake policy; each sitting is a separate purchase. The score report you receive after the exam gives your pass or fail result, your scaled score, and your percent-correct by domain. That domain breakdown is the most actionable output from a failed attempt: it tells you exactly where to focus before you rebook.

The credential is valid for 12 months from the date it is awarded. Renewal mechanics have not been published separately from the standard retake process; check Anthropic's official certification pages for the current policy before your renewal date.

As of 3 June 2026, more than 10,000 individuals hold a Claude certification across all tracks, and over 40,000 partner firms have applied to the Claude Partner Network. The programme is growing quickly, and early credential holders have a timing advantage in a market where the supply of certified developers is still thin relative to demand.

AI Skill Certs is an independent prep platform. We are not affiliated with, endorsed by, or approved by Anthropic. Our CCDV-F adaptive study, Archie tutoring, and practice exams are live today.