API
AgentPrep API
AgentPrep is a gamified coach for Anthropic's certification exams. This API is the engine behind all three surfaces — the Telegram bot, the web app and the Claude Code skill — and it is the single source of truth for question content and correctness.
Base URL: https://api.agentprep.dev. Most endpoints authenticate with a Pro license key; one real practice question is open to anyone, with no account.
Authentication
Send the license key as a bearer token:
The server is the authority on entitlement: it verifies the license is active and meters usage server-side, answering 401 when the token is missing or unknown and 403 when it is known but not entitled to that tier. Clients hold no entitlement logic. A key comes from a Pro purchase — see the landing page.
Open endpoints
No token needed. Rate limits are per client IP.
| Method | Path | What it does |
|---|---|---|
| GET | /healthz | Liveness probe. Returns {"ok":true}. |
| GET | /v1/app/sample-question | One real practice question, no account. 60 req/min. |
| POST | /v1/app/sample-question/answer | Grades an answer to that sample question. Writes nothing. |
| POST | /v1/licenses/activate | Exchanges a purchased license key for access. 20 req/min. |
Licensed endpoints
All of these require the bearer token above. This table is the map, not the contract: exact request and response shapes live in the skill file below, which is what the shipped client actually follows.
| Method | Path | What it does |
|---|---|---|
| GET | /v1/exams | Live catalog of supported certifications. Never hardcode it. |
| PUT | /v1/me/exam | Switches the caller's active certification. |
| GET | /v1/quest/today | The daily quest: today's practice questions. |
| POST | /v1/answers | Submits an answer and returns the grading + explanation. |
| GET | /v1/me/stats | XP, level, streak, per-domain mastery. |
| GET | /v1/me/readiness | Readiness score and forecast ready-by date. |
| GET | /v1/me/mistakes | The caller's missed questions, for review. |
| POST | /v1/simulado/start | Starts a 60-question / 120-minute mock exam. |
| POST | /v1/simulado/:id/finish | Closes a mock exam and returns the score breakdown. |
Discovery
Every public page here returns an RFC 8288 Link header pointing at this documentation, at the RFC 9727 catalog and at the skill — so an agent can find all three from a single HEAD /:
- /.well-known/api-catalog — the catalog, as
application/linkset+json. - /.well-known/agent-skills/agentprep/SKILL.md — the Claude Code skill: the full client contract, in Markdown.
- /sitemap.xml and /robots.txt — for crawlers.
What this API will never do
- It never calls an LLM. It is a Fastify + Postgres service. All tutoring judgment comes from the agent on the client side, using the raw material returned here.
- It never serves real exam content. Every question is original, scenario-based material written from public sources. Real exam content is protected by the Pearson VUE NDA and is not here.