Days 1–5: read + build
One domain a day, in weight order. Every day pairs reading with a hands-on exercise — like writing a minimal agent loop that cements ~8 exam facts in 40 minutes.
Week 1 of the roadmap · the hardest exam — do it first
CCAR-F is the technical anchor of the Claude certification track — the most memorization-heavy exam of the suite, testing exact API parameters, file paths, and CLI flags across five weighted domains. Here's what's on it, what isn't, and how to be ready in seven days.
01 — Exam snapshot
60 Q
1 correct of 4, multiple choice
4
Random scenarios per sitting
120 min
Closed book, online proctored
720
Pass, on a 100–1000 scale
$125
List price — discounts may apply
0
Guessing penalty — answer all
Delivered via Pearson OnVUE and booked through the Anthropic Partner Academy. The Credly badge is valid 12 months. Retakes: 14 days after the first attempt, 30 after the second, 90 after the third; max 4 attempts per rolling 12 months. Logistics reflect the official program pages as of August 2026 — verify before booking.
02 — What's on it
| Domain | Weight | ≈ Questions of 60 | What it covers |
|---|---|---|---|
| Agentic Architecture & Orchestration | 27% | ~16 | The agentic loop, coordinator/subagent design, hooks vs prompts, decomposition, escalation, sessions — the biggest domain. |
| Claude Code Config & Workflows | 20% | ~12 | Claude Code configuration, CLAUDE.md, commands, flags, and day-to-day workflows. |
| Prompt Engineering & Structured Output | 20% | ~12 | Prompts, JSON Schema, Pydantic, and getting reliable structured output. |
| Tool Design & MCP Integration | 18% | ~11 | Tool definitions, built-in tools (Read/Write/Edit/Bash/Grep/Glob), and MCP integration. |
| Context Management & Reliability | 15% | ~9 | Context windows, truncation behavior, and building agents that fail safely. |
The signature question style: a realistic production scenario, then four options that all look legitimate — one real flag, path, or parameter and three plausible fakes. CCAR-F rewards muscle memory: you either have typed it or you haven't.
Out of scope — don't study these: fine-tuning, auth/billing details, streaming/SSE, rate limits and cost math, prompt-caching internals, embeddings/vector DBs, computer use, vision, MCP server hosting, model internals.
03 — How to prepare
One domain a day, in weight order. Every day pairs reading with a hands-on exercise — like writing a minimal agent loop that cements ~8 exam facts in 40 minutes.
Take the 36-question quiz bank in exam mode, timed at ~2 minutes per question. The mock tells you whether to book.
Re-run your wrong answers, review the quick-reference tables, and book once you're scoring 85%+ — comfortably above the ~72% bar.
04 — The study kit
A 51-page, diagram-rich guide mapped to the five weighted domains of the official exam blueprint, cross-checked against Anthropic's documentation. Includes the full 7-day plan and a 36-question original quiz bank with complete answer explanations.
05 — FAQ
In our editorial opinion, CCAR-F is the hardest of the four Claude certification exams — the most memorization-heavy of the suite, testing exact API parameters, file paths, and CLI flags. The passing score is 720 on a 100–1000 scale (roughly 72%), there is no guessing penalty, and 120 minutes for 60 questions leaves time pressure low.
Five weighted domains: Agentic Architecture & Orchestration (27%), Claude Code Config & Workflows (20%), Prompt Engineering & Structured Output (20%), Tool Design & MCP Integration (18%), and Context Management & Reliability (15%). The technology checklist covers the Claude Agent SDK, MCP, Claude Code, the Claude API, the Message Batches API, JSON Schema, Pydantic, CLAUDE.md, and the built-in tools.
Per the published objectives: fine-tuning, auth/billing details, streaming/SSE, rate limits and cost math, prompt-caching internals, embeddings and vector databases, computer use, vision, MCP server hosting, and model internals. Don't spend study time there.
We suggest taking it first: it's the hardest exam, and its content echoes through both other technical exams (CCDV-F and CCAR-P), making each subsequent exam easier. If you already have real generative-AI architecture experience, you could go straight to CCAR-P.
06 — Next step
Week 2 of the roadmap is CCDV-F — the natural follow-on. Most of the hard material is already in your head; this exam turns it toward implementation.
Independence. This is an independent, unofficial study resource — not affiliated with, endorsed by, sponsored by, or approved by Anthropic, PBC or Pearson Education, Inc. “Claude,” the certification names, and the exam codes are trademarks of Anthropic, PBC, used here solely to identify the exam this page covers. Exam logistics reflect the official program pages as of August 2026 and can change — verify before booking. Every practice question in our guides is original and written to the publicly published exam objectives.