Claude Code, in the work you already do
Short lessons on context, permissions, MCP and automation — 90 seconds of theory, then one real decision with every alternative explained. The docs tell you what exists. Here someone checks whether you got it right, and says why when you didn't.
Quick challenge · no signup
Pick an answer, get instant feedback, keep moving. The two toughest challenges stay locked until the finish.
This sprint needs JavaScript, and right now it isn’t running. The questions below will not respond — but the quiz is not the only way in. Take it on Telegram: question 5, the answer key and 3 new questions a day → Or continue free in the app →
Question 1 of 5
An engineering lead is evaluating Claude Code for a monorepo with roughly two million lines of proprietary code. Two concerns are blocking approval: whether the codebase must first be uploaded and pre-indexed — embedded into some vector database — before Claude can work with it, and whether a repo that large can even fit alongside a conversation in the model's context window.
How does Claude Code actually gather context about a codebase this size?
Question 2 of 5
A platform team maintains three hand-rolled integrations of the same internal knowledge base: a bespoke plugin for their in-house agent, a custom extension for the engineers' AI coding tool, and another for the support team's AI desktop assistant. Every schema change to the knowledge base means re-updating all three, and a fourth AI surface is about to launch.
Architecturally, what changes if the team instead exposes the knowledge base through a single MCP server?
Question 3 of 5
After enabling extended thinking on a financial-analysis endpoint, a team's cost dashboard shows per-request spend roughly tripling — yet the answers users see are about the same length as before, and the thinking summary included in responses is only a short paragraph. Separately, a few complex requests now come back truncated with stop_reason: "max_tokens", which never happened before the change.
What single fact about extended thinking explains both observations?
Sprint complete
0/3
Prefer it by email?
Done — it's on its way to your inbox.
That didn’t go through — check your connection and try again.
The point
Not "you'll have read about it". Four decisions you'll be making in your own work by the end of the week.
How it works — three pieces, one score.
Your daily quest — 3 questions free, 5 on Pro — right in Telegram or the browser. Never trivia: each one is a real situation where two options look defensible. Every answer comes back with the full reasoning and a link to the official docs, on any plan. Miss a day and only your streak resets — your progress stays exactly where you left it.
Open the web app and see all 30 lessons laid out across 5 domains — 90 seconds of theory each, then the decision that theory was for. Work through them in order, clear a checkpoint to unlock the next one, and always know exactly where you stand.
A 20-question placement test calibrates where you actually stand — free to take, full results on Pro. From there every answer builds your mastery score domain by domain, so "I think I know this" turns into a number you can check. The arithmetic is below, in the open.
No account needed to start learning: how the agentic loop works (27% of the exam) and agent permissions and hooks (20%) are free to read right now — the two subjects that decide most of the work, and the two heaviest domains on the test.
No black box
This is the whole model — what the bot and the skill actually compute after every answer. No hidden weighting, no vibes.
No LLM grades you behind the scenes. Every number above, you could compute yourself from your own answer log.
Before you open a single domain
Twenty questions, every domain of your certification represented, no feedback until the end — answer it the way you would on exam day. What comes back: a level (Foundations, Intermediate or Advanced), a score out of 1000, and exactly which domain to fix first.
Part of Pro. Create your free account, then unlock the diagnostic.
Certifications
Full prep for Anthropic's certifications stays in the same subscription: mastery domain by domain, your ready-by date, and every certification included. Two things said plainly before you pay for anything.
First: sitting the exam requires a work email from a company in the Claude Partner Network. A personal address does not get you registered — that gate is Anthropic's, not ours, and it is worth checking before you buy anything anywhere. Second: the exam tests exactly the skills in these tracks, which is why the lessons come first here and the exam comes after.
Every new Anthropic certification: included, same subscription.
Pricing
The lessons, the theory and the explanation behind every answer stay open on the free plan. Pro pays for the reps and the measurement: the full quest, unlimited practice, every domain scored, and the date you'll be ready.
$0free, forever
Runs in your browser. The Telegram bot is free and needs no Claude subscription.
$49.90$24.95per year
or $4.99 a week — cancel anytime
Secure checkout by Stripe. License key by email, refunds unconditional within 7 days — just ask. Both plans are subscriptions and renew automatically until you cancel. Terms, renewal & cancellation.
Behind AgentPrep is Marcelo Mendes, an AI product lead who works with agents and LLMs in production. The free plan is permanent and never asks for a card; Pro cancels whenever you want, from the Account screen — and your progress is never deleted.
Questions before you buy? Write to contato@agentprep.dev.
| Capability | Free | Pro |
|---|---|---|
| Daily practice quest | 3 of 5 a day | All 5 |
| Explanation + official-doc links after each answer | ||
| Extra practice beyond your daily quest (Telegram) | Unlimited | |
| Mastery % and stars per domain | All on Telegram · weakest in the app | Both, every domain |
| Your exact ready-by date | Locked | Full date, with the math |
| Theory cards (concept explainers) | Every domain | Every domain |
| Exams included | All 4, switch anytime | All 4, switch anytime |
| Recent-mistakes review | ||
| Placement test (20 questions) | Results locked | Full results |
FAQ
Yes. It runs in your browser — sign in and start: 3 quest questions a day, the full 30-lesson trail, your overall mastery, no card and no key. The Telegram bot is free too, and needs no Claude subscription. Pro ($24.95/yr) unlocks the full 5-question quest, unlimited extra practice on Telegram, full mastery and stars for every domain, and your exact ready-by date.
No. AgentPrep runs entirely in your browser and on Telegram — no Claude subscription, API key or install required, on Free or Pro.
No — and they can't be. Every question is 100% original and scenario-based, written from public material: official exam guides, Anthropic docs, Academy content. Using real exam questions would violate Pearson VUE's NDA. We don't, on principle.
It's continuously updated to track the exam and the Claude ecosystem as both change — not a fixed pile you empty out. Readiness isn't "you cleared the bank," it's your mastery holding 80+ in every domain. That's what §04 actually measures.
Anytime, at no cost. One subscription covers every certification we support — pick one to start, switch whenever you want, and new ones join the same subscription the day they ship.
No. There is a weekly plan you can cancel whenever — §07 has both prices. It is the right call when the exam is close. Past a few months of preparing, the year costs less than the same weeks would, and it is the same product either way. Both are subscriptions and renew automatically — the year every year, the week every week — until you cancel. Cancelling from your Account screen stops future renewals, and you keep Pro until the end of the period you already paid for.
A license key lands in your inbox right after checkout — paste it into the bot or the skill, either one activates both. Refunds: unconditional within 7 days — just ask, no reason needed.
Start free today. Go Pro — $24.95 a year, or $4.99 a week if the exam is close — when you want the full quest, full mastery in every domain, and your exact ready date.