Claude Certified Developer Foundations Build & Pass CCDV-F

What you'll learn

---

Predict Claude's behaviour from mechanism: tokens, the context window, sampling and non-determinism
Diagnose a failing prompt against four symptoms instead of adding words until it works
Move output control from the prompt into the API with structured outputs, and know its real costs
Build a tool-use loop with schemas Claude selects correctly, and know which step your code owns
Consume a stream without corrupting state when it drops mid-message
Engineer context deliberately: what enters, what returns as a summary, what never enters
Decide workflow or agent before line one, and wire a loop with exit conditions and a human gate
Operate Claude Code's six permission modes, settings hierarchy, hooks and subagents correctly
Build and configure MCP servers, choosing transport and scope without leaking credentials
Define done with evals, match graders to output shape, and calibrate a judge that is defensible
Handle failure properly: retriable versus terminal, and the HTTP 200s that are not usable answers
Defend against prompt injection with capability limits, hooks and containment rather than wording
Package a build as a reusable accelerator and contribute it back so a maintainer can accept it
Choose a platform, pin a version, and map trust boundaries that survive a regulated review
Requirements

---

Comfortable writing code in Python or TypeScript
Familiarity with REST APIs and JSON
An Anthropic API key for the hands-on parts (a small amount of credit is enough)
No prior experience with Claude or LLM APIs is assumed
Description

---

This course contains the use of artificial intelligence.
Most Claude courses teach you the API surface. This one teaches you which decision prevents which production failure.
Every section follows one spine: a Claude integration that worked well in development now has to hold up in production. Longer sessions, larger tool outputs, interrupted streams, tighter cost limits, and actions that cannot be undone. Twenty-two lectures open on a real incident - an agent that edited a production file, a stream that left a half-written tool call in the history, a fetched page that gave the orders - then diagnose it and build the mechanism that would have prevented it.
You will learn tokens, the context window and sampling as behaviour you can predict; prompting craft and structured outputs; the tool-use loop and schema design; streaming without corrupting state; context engineering; agent construction and memory scope; Claude Code's permission model, hooks, subagents and MCP servers; evals, graders and calibrated judges; failure handling, cost levers and prompt injection defence; and finally packaging, contributing back, deployment and trust boundaries.
Every technical specific was verified against current documentation at recording time. Where the source material was wrong - and in several places it was - the correction is named rather than quietly patched, so you can see how to check these things yourself.
Includes 5 section quizzes, 11 graded debugging assignments with full model answers, and a 53-question final practice exam weighted to the published CCDV-F blueprint.
This course is independent preparation for the Claude Certified Developer - Foundations exam. It is not official, not endorsed by Anthropic, and not affiliated with them. Completing this course does not award the credential; learners register and sit the exam themselves.
This course's narration is synthesised and its visuals are programmatically generated.
Who this course is for

---

Developers building production applications on Claude rather than prototypes
Engineers preparing for the Claude Certified Developer - Foundations exam
Teams who have shipped an LLM feature and been surprised by cost, latency or a silent failure
Anyone who has to defend an AI integration in a security or procurement review
Developers moving from prompt experimentation to agents with tools and real consequences
Homepage

---