Claude Certified Developer Foundations (CCDV-F) Exam Guide (2026)

If you ship a Claude application against the Messages API, wire a tool or an MCP server, and decide when a batch job beats a realtime call, you are in the right exam.

CCDV-F is Anthropic's Developer Foundations sitting. It is not Associate Foundations. It is not Architect Foundations. It is not Architect Professional. The official exam guide names the audience as technical people who build, integrate, and ship production Claude applications, agents, and workflows.

This page covers the current Version 1.0 blueprint, effective July 2026. Weights and task names below are copied from that PDF.

Who this exam is for

Take it if the work already looks like production code. You send messages through the Claude API or a client SDK. You choose streaming, caching, or the Message Batches API from cost and latency. You write a tool schema that another service can call. You decide whether a reusable capability belongs in a custom tool, a Skill, or an MCP server. You pin a model version. You keep secrets out of the prompt. You isolate a failure in the integration layer from a bad model output.

The official guide names AI and machine learning engineers, technical leads, and senior software engineers. These people sit between a requirement and a running system. They build agents and workflows with the Claude Agent SDK and agentic frameworks. They operate Claude Code on a repository. They write prompts and manage context. They run a simple eval. They apply a guardrail.

The minimally qualified candidate already ships Claude-powered applications, agents, and workflows. That person has moved past a chat window. They translate a technical requirement into working code. They feel the limits: tokens, non-determinism, untrusted tool output, and a context window that fills.

Skip it if the hard part is a daily Claude Project, a prompt you iterate, and an output you stand behind. That sitting is Associate Foundations. Skip it if the hard part is whether a hook must block a refund, whether a coordinator should spawn a subagent, or whether plan mode beats a one-function Claude Code edit. That sitting is Architect Foundations. Skip it if the hard part is enterprise retrieval, a named regulation, or a stakeholder lifecycle. That sitting is Architect Professional.

Recommended experience is one to five years of software engineering and at least six months with Claude or a comparable LLM system. Python or TypeScript, REST APIs, and CLI fluency are the working tools. None of that is a gate. The PDF states there are no mandatory prerequisites or required courses. The credential is awarded from exam performance alone.

How the exam works

You cannot buy a seat with a personal Gmail or Outlook address. The Partner Academy FAQ says registration needs a Claude Partner Network company email on a recognized domain. Personal email is rejected. If you sign in and see "You aren't signed in with a partner company employee email," fix the domain first. Adding a domain to a partner record takes 7 to 10 days. You must be 18 or older. Age is checked at Pearson against a government ID.

Register and pay on the Developer Foundations certification page in Anthropic Partner Academy. Checkout shows any partner-tier discount that applies. Pearson then emails the scheduling link. Pick online proctoring or a Pearson test center. After you pay, the registration stays valid for 5 years.

The official CCDV-F exam guide, Version 1.0, sets the sitting.

Scoring is criterion-referenced. You pass by meeting a fixed standard, not by beating other candidates. Domain percents on the score report are study feedback. They do not create a second pass bar.

The exam is English only and closed book. Browser translation tools are not allowed under Pearson proctoring. Notes, docs, and AI assistants are also out. On exam day the workspace stays clear. You accept a confidentiality agreement before the items start. Declining it ends the session with no refund.

The name on the Pearson profile must match the government ID exactly. Ask for a name correction at certifications-support@anthropic.com before you schedule. The exam guide for this sitting says you may cancel or reschedule up to 24 hours before the appointment. A change inside 24 hours forfeits the fee. A no-show or a late arrival also forfeits the fee.

Failed attempts wait 14 days, then 30, then 90. You may sit the same exam four times in a rolling 12 months. The fee applies each time.

The badge lasts 12 months. Credly by Pearson issues it. On-time renewal is a free, non-proctored Partner Academy assessment. A lapsed credential requires the full exam at the full fee. If the blueprint changes a lot, Anthropic may require the full exam instead of the renewal assessment.

Partner Network tier credit counts for Developer Foundations. That is an official program rule on the hub and in the FAQ. Architect Foundations and Architect Professional also count. Associate Foundations does not. Sit this exam for the Developer skill. The tier credit is a program fact, not the reason to pick the wrong role.

Domain weights

The blueprint has eight domains. Weights come from the job task analysis in the official guide.

Domain 2 is a third of the scored items. If you only memorize Claude Code flags and skip the Messages API, the math is against you. Domain 5 is 16.8 percent. A week that never touches model choice, thinking modes, or caching will show up on the score report. Domain 1 is 14.7 percent. Agents are real work on this sitting, not a side topic.

Domains 2, 5, and 1 together are about two thirds of the sitting. A month that ships one integration, picks a model on purpose, and builds one agent loop covers those three at once. That is also the prep path the official guide describes.

Claude Code at 3.1 percent and Eval at 2.6 percent are real. They are small. Do not spend the month only in the CLI or only on an eval dashboard.

How it differs from Associate and Architect

Developer Foundations shares a Claude badge family with Associate and the Architect exams. It does not share a blueprint.

The Partner Academy hub says to pick a role, start at Foundations, then advance to Professional where that path exists. Associate is the productivity-tool credential. Developer is the engineer credential for people who build with the Claude API, Claude Code, and MCP. Architect Foundations is the first technical Architect sitting. Architect Professional is the enterprise design and governance sitting. Foundations does not become Professional by renewal. You can sit a Developer or Architect exam without holding Associate.

CCDV-F, from its own Version 1.0 guide, is a 53-item sitting at $125. It uses eight domains. The tasks sit on API mechanics, application design, agents, model choice, prompts and context, tools and MCP, and security. Partner Network tier credit applies.

CCAO-F is a 60-item sitting at $99. It uses seven domains. The tasks sit on prompts, output judgment, product and model choice, workflow design, Project configuration, policy, and troubleshooting. Partner Network tier credit does not apply.

CCAR-F is a 60-item sitting at $125. It uses five domains. A sitting draws 4 scenarios from a published bank of 6. The tasks sit on agentic loops, coordinator and subagent design, hooks, MCP tool interfaces, CLAUDE.md hierarchy, plan mode versus a direct Claude Code run, structured output, and context reliability. It counts toward partner tier.

CCAR-P is a 63-item sitting at $175. It uses seven domains. The Professional PDF does not publish a scenario bank. The weight sits on enterprise integration, evaluation programs, named compliance, stakeholder lifecycle, and business value. It counts toward partner tier.

A useful split when you pick a seat:

• If the hard part is a daily Claude Project, a prompt you can iterate, and an output you can stand behind, sit Associate.
• If the hard part is a Messages API integration, a tool schema, a batch job, or an MCP server you can reuse, sit Developer.
• If the hard part is whether a hook must block a refund, whether a coordinator should spawn a subagent, or whether plan mode beats a one-function edit, sit Architect Foundations.
• If the hard part is whether the agent has too many tools, whether retrieval broke after a document refresh, or whether a control satisfies a named regulation, sit Architect Professional.

Do not grind an Associate or Architect bank and expect the Developer items to feel the same. The Developer sample items in section 8 already show a different habit. They ask whether an overnight document job belongs on the Message Batches API, whether a hidden instruction in a web page is untrusted input, and whether a shared inventory service should be an MCP server.

How to study for the live blueprint

The official guide says there is no single required course. Anthropic does not promise that any resource produces a pass. The same section tells you to combine hands-on work with the blueprint, official docs, one real application, developer competencies, and the sample items.

Week 1. Ship one integration. Domain 2 is 33.1 percent. Build a small Claude application that calls the Messages API from Python or TypeScript. Send a tool. Stream one response. Pin the model ID. Store the key outside the repo. Write the requirement in one sentence before you write the call. Then add a second path that does not need an instant reply.

Official docs to keep open this week: the Messages API, the client SDKs, and batch processing. The official Domain 2 sample already wants the Message Batches API for 10,000 overnight documents when cost matters and morning is soon enough. Parallel synchronous calls do not cut per-token cost. Lowering max_tokens or shrinking the model does not answer the batch-versus-realtime question.

Keep CLAUDE.md and settings.json in the same week. Domain 2 also scores configuration management and how Claude reads instructions across Claude Code, Desktop, claude.ai, the API, and SDKs. A settings file you can reopen next week beats a pile of screenshots.

Week 2. Model choice and context. Domain 5 is 16.8 percent. Domain 6 is 11.0 percent. Pick Opus, Sonnet, or Haiku from quality, latency, and cost, the way the PDF names those tiers. Read the current models overview for today's IDs, effort levels, and thinking modes. Do not treat the most capable model as the default.

Turn on prompt caching when the prefix is stable. Official prompt caching docs are the reference. Track tokens on one real job and write the cost in one line. When context fills, prune tool output or isolate the next step in a subagent. Do not keep stuffing a dead thread.

Write one prompt you can score twice. Put the constraint in the system message. Put the untrusted payload in the user turn or a tool result. Parse the structured output as if it might be wrong.

Week 3. Agents, tools, and MCP. Domain 1 is 14.7 percent. Domain 8 is 10.6 percent. Build one agent loop with the Claude Agent SDK or a custom loop you can explain. Decide whether the job is a fixed workflow or an agent that must pick tools. Add a hook that blocks one destructive action.

Then expose one internal REST operation as a tool. If two applications need that same capability, put it on an MCP server. The official Domain 8 sample already wants that shape for an inventory service. Hard-coding the logic into a system prompt is not reusable. Pasting the table into context is not live access. A built-in tool does not reach an arbitrary internal API.

Official docs for this week: tool use and the MCP overview.

Week 4. Security, the thin domains, and the official samples. Walk one user-submitted web page through the jailbreak and prompt-injection guidance. The official Domain 7 sample wants untrusted page content kept away from trusted instructions, with a guardrail or hook so an injected line cannot call a sensitive tool. A polite sentence in the system prompt is not that control. A larger model is not that control.

Give Claude Code one real repository hour. Read the Claude Code overview. Initialize the repo. Write a CLAUDE.md a teammate can follow. Try a Skill or a slash command. Then stop. Domain 3 is 3.1 percent.

Break one integration on purpose. Read the trace. Write whether the fault sat in your code or in the model output. Then stop. Domain 4 is 2.6 percent.

Work the three official sample items in section 8 for format. They are illustrative. They are not the live bank. After each one, write the constraint in one sentence before you look at the rationale. If you can pick the answer from a product name alone, you are practicing the wrong habit.

The official Partner Academy prep path for this exam walks the same ground: model and SDK foundations, production prompts and agent loops, Claude Code and MCP, then evals and security. Use it as structured reading. Use the PDF as the outline you are scored on.

Where this exam stops

This sitting tests whether you can build, integrate, and ship a Claude application at a foundational level. It does not test whether you can run Claude as a daily productivity tool. It does not test whether you memorized Architect domain weights.

Stay on the Developer blueprint.

• An overnight job of 10,000 documents is a Domain 2 item. The Message Batches API is the fit. Parallel realtime calls are the waste.
• A hidden instruction in a fetched page is a Domain 7 item. Isolate the page from trusted instructions. A prompt that asks the model to behave is not a control.
• A shared inventory REST API is a Domain 8 item. An MCP server is the reusable fit. A built-in tool is not a private API client.
• Daily Project work, citation checks, and knowledge-source hygiene belong to Associate. Escalate that work. Do not invent a Developer answer for a Project setting.
• Coordinator design, plan mode versus a one-function edit, and a hook that must never miss belong to Architect Foundations when those are the hard part of the job.
• Named regulations, stakeholder lifecycle, and firm-wide evaluation programs belong to Architect Professional.
• Academy course quizzes on academy.claude.com are not this Pearson sitting. Study the exam guide PDF.
• A personal email that can open Claude.ai still cannot register for this exam.
• Claude Code is real and small. Do not turn the month into a CLI workshop.
• Eval is real and small. A single trace you can explain is enough for Domain 4. Do not starve Applications to polish a dashboard.
• Fine-tuning, embeddings, and cloud-provider setup are not this job. If the stem wants a trained model, you are practicing the wrong exam.
• If you can delete the API call, the tool schema, and the batch-versus-realtime choice, and still pick the answer from a product name, you are studying a slogan. This sitting scores the integration.

The official guide also tells you what prep is not. No course, including the Partner Academy path, is required. No resource is guaranteed. If a study note cannot point at a domain and a task in section 6, drop it.

How this maps to CloudFluently

Study with the Claude Certified Developer Foundations practice exam sets. Use this page for how the Developer sitting is built and how it parts from Associate and Architect. Treat the Claude Certified Developer Foundations exam guide PDF as the outline the sitting is scored on.

Use the live sibling courses when the next gap matches those blueprints. Study with the Claude Certified Architect Foundations practice exam sets when the work is agents, tools, MCP, and Claude Code from the Architect side. Use the Claude Certified Architect Professional practice exam sets when the next gap is enterprise integration, evaluation, and governance. Use the Claude Certified Associate Foundations practice exam sets when the next gap is daily Claude Projects, output judgment, and policy.

Official docs for the features Domain 2, Domain 5, and Domain 8 name: the Messages API, the models overview, tool use, and batch processing.

Official links

Official exam guide: Claude Certified Developer Foundations exam guide PDF. Official prep: Developer Foundations prep course. Register from Developer Foundations certification after your partner domain works. The four-exam catalog lives on the Partner certifications hub. Program rules live on the certification FAQ. Pearson support for this program is on Pearson VUE for Anthropic.

Frequently asked questions

Is CCDV-F the same exam as Architect Foundations? No. Developer Foundations is the engineer sitting for people who build, integrate, and ship Claude applications. CCAR-F is the architect sitting for production applications with the Agent SDK, Claude Code, the API, and MCP. Both count toward partner tier. The blueprints differ.

Is CCDV-F the same exam as Associate Foundations? No. Associate Foundations is the productivity-tool credential and does not count toward partner tier. Developer Foundations is the API and application sitting and does count.

Do you need another Claude certification first? No. There is no prerequisite. You can sit Developer without Associate. Anthropic still calls Foundations the natural start inside each role.

Can you register with a personal email? No. Registration needs a Claude Partner Network company email on a recognized domain. Personal email is rejected. Domain adds take 7 to 10 days.

How many questions are there, and how long is the seat? Fifty-three items. 120 minutes of questions. About 135 minutes with check-in.

What is the pass mark? A scaled 720 on a 100 to 1,000 scale. You pass the exam as a whole. Domain percents on the score report are study hints.

How long does the badge last? Twelve months from the award date. On-time renewal is a free open-book Partner Academy assessment. A lapsed credential requires the full exam again.

Does this exam count toward Partner Network tier? Yes. Developer Foundations counts. That is the official hub and FAQ rule. Architect exams also count. Associate Foundations does not.

What should you study first? Applications and Integration. Domain 2 is 33.1 percent. Then model selection, then agents and workflows.

Where is the official outline? Claude Certified Developer Foundations exam guide PDF. Register and book from Developer Foundations certification. Use Anthropic for task statements and weights. Use this page for how to study and how Developer parts from CCAO-F, CCAR-F, and CCAR-P.