Anthropic · CCDV-F · Foundations
Claude Certified Developer – Foundations (CCDV-F)
Exam update: Anthropic has announced that an updated exam guide will be available from 17 November 2026 and that the exam will include updated content from 7 December 2026. Taking the exam on or after 7 December 2026? Study from the updated guide. Taking it before then? Study from the current guide (v1.0). We'll update this page once the new guide is out.
Technical professionals who build, integrate and ship production Claude apps, agents and workflows: AI/ML engineers, technical leads, senior software engineers. Not for non-technical/casual users or prompt-writing-only roles.
60 free practice questions · full timed mock exam · study plan
Exam facts
- How many questions are on the CCDV-F exam? 53 questions, multiple choice and multiple response.
- How long is the CCDV-F exam? 120 minutes.
- What is the CCDV-F pass mark? 720 scaled (100-1,000)
- How much does the CCDV-F exam cost? $125 USD before any partner discount.
- Who can take the CCDV-F exam? Only people who work at Claude Partner Network organisations can take the exams, registering with a company email on a domain listed on their organisation's partner record (personal email addresses won't work). You must be 18 or over and show a valid, unexpired, government-issued photo ID whose name exactly matches your exam profile.
- How is the CCDV-F exam delivered? Proctored by Pearson VUE: online proctored (OnVUE) or Pearson test center; registration via Anthropic Partner Academy (Claude Partner Network members, 18+, partner email domain)
- How long is the CCDV-F certification valid? Valid for 12 months from the day you earn it. Renew for free before it expires by passing an open-book, non-proctored assessment on Partner Academy; if it lapses, you must pass the full exam again at the full fee.
Checked 2026-11-15 against the official exam guide (Version 1.0, effective July 2026 (initial publication July 2026)). Official exam guide · Official certification FAQ
Retakes, rescheduling and more
- How long should I set aside on exam day? The exam itself is 120 minutes, but Anthropic advises planning for about 135 minutes of total seat time, to cover check-in, instructions and a short survey afterwards. For an online (OnVUE) exam, Pearson VUE also asks you to start check-in 30 minutes before your appointment time, for system checks, ID photos and a room scan, so allow time beyond the exam itself.
- Can I take the exam online or at a test centre? Either. All four exams are delivered by Pearson VUE, and you choose between OnVUE online proctoring and a Pearson test centre at the time you schedule. To switch, cancel and book again. If your government ID is from Belarus, Cuba, North Korea, Russia, Syria or a restricted region of Ukraine, you can't use OnVUE and must test at a test centre.
- How do I register and book the exam? Since 12 November 2026 you register in Anthropic Partner Academy, go to Pearson VUE through the Partner Academy sign-in, then pay and pick a date at Pearson's checkout, where any partner discount is applied automatically. You need a company email on a domain listed on your organisation's Claude Partner Network record. Personal email addresses won't work.
- What do I need on exam day? A valid, unexpired, government-issued photo ID. The name on it must exactly match your exam profile, or you can't test and you lose the fee. You must be 18 or over, and your age is checked against the ID.
- Is there a discount on the exam fee? It depends on your organisation's Claude Partner Network tier. Registered-tier partners pay full price. Select, Preferred and Global Premier partners get 50% off, applied automatically at Pearson's checkout. Up to and including 31 December 2026, Global Premier partners get 100% off instead, so their exams are free; after that date, the standard 50% discount applies. The discount is locked in when you schedule, so book by 31 December 2026 to keep the 100% discount.
- Why do some sites say Architect – Foundations (CCAR-F) costs $99? That was the old CCAR-F price. From 30 June 2026, the same day exam delivery moved to Pearson VUE, the CCAR-F fee went up from $99 to $125 (USD, before any partner discount). Registrations completed before that date paid $99.
- Is the exam open book? No. The exams are closed book: no notes, documentation, translation tools or AI assistants. Only the separate renewal assessment is open book.
- What languages is the exam in? English and Japanese only. You pick the language when you schedule at Pearson. To switch a booked English exam to Japanese, cancel it and book again through Partner Academy. No translation tools, devices or software, including browser translation, are allowed during the exam; if English is your second language, you can ask Pearson for extra time instead.
- What happens if I fail? How do retakes work? You wait 14 days after a first failed attempt, 30 days after a second and 90 days after a third. You can sit the same exam at most four times in any rolling 12-month period. These limits apply to each exam separately, and Anthropic's policies say they reset when an exam moves to a new version. Each retake costs the full exam fee, and any partner discount you get applies to retakes too. Once you pass, you can't retake the exam to raise your score.
- Can I cancel or reschedule the exam? Test-centre exams can be cancelled or rescheduled free up to 48 hours before. For online (OnVUE) exams Anthropic says up to one hour before, but Pearson VUE's FAQ says 48 hours, so make changes at least 48 hours ahead to be safe. The 24-hour cutoff in the July 2026 exam guides is out of date. If you change after the deadline or don't show up, you lose the fee and must pay for a new attempt; the exam guides say the same applies if you arrive after the late-arrival window. If you paid at Pearson's checkout and cancel in time, the refund starts automatically and takes 7 to 10 business days; rescheduling instead keeps your locked price and discount.
- How does renewal work? Each certification is valid for 12 months from the day you earn it. Before it expires, you renew for free by passing an open-book, non-proctored online assessment on Partner Academy, which you can retake as many times as you need. Passing adds 12 months to your current expiry date, not to the day you renew, so renewing early does not shorten your validity. If it lapses, you must pass the full exam again at the full fee. If Anthropic substantially updates an exam, it may require holders to pass the current version of the exam, rather than the renewal assessment, to stay certified; Anthropic decides this case by case. Anthropic hasn't yet said how early the renewal window opens. Certifications earned before 30 June 2026, when validity was 6 months, were automatically extended to 12 months from the date they were originally earned.
- How do I get my certificate or badge? After you pass, Credly, Anthropic's digital badge platform, emails you an invitation to accept your badge. You can then share it on LinkedIn and download a PDF certificate.
- Which certifications count toward Claude Partner Network eligibility? Developer – Foundations (CCDV-F), Architect – Foundations (CCAR-F) and Architect – Professional (CCAR-P) count toward Claude Partner Network eligibility. Associate – Foundations (CCAO-F) does not.
- Can I take the exam from any country? Only from the countries and regions Anthropic supports (see its Supported Regions Policy). Crimea, Donetsk, Kherson, Luhansk and Zaporizhzhia are excluded, and Pearson VUE has suspended exam delivery to residents of Iran since 8 September 2026.
What the exam covers
D1 Agents and Workflows (14.7% of the exam)
- Agent Architecture (4.5%): workflow-vs-agent decision criteria, manager/supervisor hierarchies, role of subagents
- Agent Construction with Claude (5.3%): Claude Agent SDK, custom agent loops/harnesses, self-hosted vs Anthropic-hosted managed agents, hooks for deterministic actions
- Agent Patterns and Frameworks (4.9%): tool-use loops, sub-agents, memory, context-window management; frameworks such as Strands, LangGraph, PydanticAI
D2 Applications and Integration (33.1% of the exam)
- Understanding Requirements (3.4%): functional and infrastructure requirements from business needs and solution architecture
- Systems Life Cycle (2.8%): concepts for developing, operating and maintaining IT systems
- Claude API Mechanics (6.8%): messages, tools, streaming, vision, thinking, caching, third-party vendor access, Messages API data access, Batch API, realtime vs batch trade-offs
- Software Engineering Foundations (7.4%): REST, JSON, async programming, version control, SDLC, code review, refactoring
- Claude Application Design (8.6%): how Claude reads instructions across Claude Code/Desktop/claude.ai/API/SDKs, content boundaries, schema design, session hygiene, plugin management
- Configuration Management (4.1%): CLAUDE.md, settings.json, model version pinning, prompt versioning, plugin dependencies
D3 Claude Code (3.1% of the exam)
- Claude Code Operation (3.1%): Rules, Skills, Commands, Agents, Agent Memory; session management, built-in/custom slash commands, headless, streaming and auto mode; CLAUDE.md hierarchy, repo init, settings.json
D4 Eval, Testing, and Debugging (2.6% of the exam)
- Debugging and Error Handling (2.6%): identify error types, choose recovery strategy, trace analysis of failure modes, isolate integration-layer vs model-output problems
D5 Model Selection and Optimization (16.8% of the exam)
- LLM Fundamentals (5.2%): tokens, context windows, sampling, non-determinism; fast mode, extended/adaptive thinking, effort levels; zero/single/multi-shot prompting
- Technical Fundamentals (6.1%): engineering basics such as SDKs wrapping REST APIs, websockets
- Model Selection and Tradeoffs (2.7%): Opus vs Sonnet vs Haiku, quality/latency/cost trade-offs, breaking changes across model releases
- Cost and Token Management (2.8%): token budgeting/tracking, cost modeling, prompt caching and cache checkpoints
D6 Prompt and Context Engineering (11% of the exam)
- Context Engineering (3.8%): context window management, preventing drift/bloat (tool output pruning, compaction), isolation via subagents/multi-step workflows
- Prompt Engineering (4.6%): clarity, few-shot, system vs user placement, output constraints, iterative refinement, input sanitization
- Output Handling (2.6%): structured output, response validation, defensive parsing, skepticism of confident output
D7 Security and Safety (8.1% of the exam)
- AI Application Security (3.2%): prompt injection, jailbreak defense, untrusted input, data leakage, PII, authn/authz, confidentiality and integrity
- Guardrails and Safe Deployment (2.3%): content policy, layered guardrails, secure-by-design (privacy, IAM, least privilege)
- Claude Hooks (1.0%): hooks as guardrails against destructive actions
- Identity, Secrets, and Key Management (1.6%): secrets/API keys across dev and prod, identity validation, access approval, access monitoring
D8 Tools and MCPs (10.6% of the exam)
- Tool Implementation (4.4%): tool use/function calling, tool descriptions, error handling, client- vs server-side tools, approval patterns, tool set design
- MCP Server Development (2.1%): authoring/deploying MCP servers, resources/tools/prompts, transports (stdio, sockets), client vs server
- Agentic Customization (4.1%): choosing among built-in tools, custom tools, Skills and MCPs
Sample practice questions
D1
An insurer processes claim forms. Every claim goes through the same three steps: extract fields, check them against the policy database, then draft a decision letter. Auditors need each step's output inspected before the next runs. Which design fits best?
- An autonomous agent given all three capabilities as tools and left to choose the order
- One large prompt that performs all three steps at once and returns the final letter
- A coordinator agent that spawns a separate subagent for each claim field it extracts
- A prompt-chained workflow with a programmatic validation check between steps
D1
A research assistant writes market reports. The subtopics worth investigating depend on each question and cannot be listed in advance; results must be merged into one report. Which pattern is the best fit?
- Prompt chaining: a fixed sequence of calls that each handle one predefined report section
- Orchestrator-workers: a lead model breaks down the task, delegates subtasks to workers, then synthesizes
- Routing: classify each question once and send it to a single specialized prompt for the whole of the report
- Parallel voting: run the same full report prompt several times and keep the most common answer
D1
A coordinator agent finds that a bug only reproduces on Postgres 16, then delegates a fix to a subagent. The subagent returns a generic fix that ignores the Postgres detail. What is the most likely cause and fix?
- The subagent never saw the parent's history; pass the finding, paths and constraints in its prompt
- The subagent's model was too small; switch it to a more capable model and retry the same delegation
- The subagent ran out of turns; raise its maxTurns so it has time to rediscover the database detail
- The subagent lacked tools; give it the same full tool set as the coordinator so its access matches the parent's exactly