Learn the judgment the exam measures by building the protocol the exam describes.
Status: Local preview Guide version: 1.0 Guide effective date: September 2026 Last verified: 2026-09-24
This free curriculum prepares you for the Model Context Protocol Associate (MCPA) exam from the Agentic AI Foundation, delivered through Linux Foundation Training and Certification:
| Exam | Credential | Level | Time | Fee | Core route |
|---|---|---|---|---|---|
| MCPA | Model Context Protocol Associate | Beginner | 90 min | $250 | 34 lessons |
The exam is online proctored, multiple choice, aligned to the Model Context Protocol specification dated 2026-07-28, valid for two years, with one retake included and a twelve-month eligibility window. The official exam item count and passing score are not published, so this curriculum's three 60-question mocks are original practice sets, not the official length, and practice percentages cannot predict an official outcome. The MCPA page lists 90 minutes; the launch announcement stated 120 minutes. Confirm current pricing, format, duration, and eligibility on the official page before scheduling, because program details can change. Every exam fact and its retrieval date is recorded in research/source-verification-ledger.md.
This curriculum is AI-native. Claude Code, Codex, ChatGPT, Cursor, or another agent can teach the route, run the checked-in lab, review the artifact you build, administer the lesson quiz, and resume from saved progress.
Start with the GitHub learner guide, or install the portable certification tutor skill at ../../skills/mcpa-certification/SKILL.md:
npx skills add rohitg00/ai-engineering-from-scratchThen ask your agent to run:
/mcpa-certification
A local Claude Code session discovers the same skill from .claude/skills/
after you clone the repository. Harnesses without slash-command support can read
GETTING_STARTED.md and the tutor skill directly. Learner progress lives in
MCPA-CERTIFICATION.md; learner work lives under learning-artifacts/mcpa/.
The checked-in outputs/ files remain reference artifacts and are never
overwritten. This curriculum lives outside the EPUB/PDF book workflow and is not
converted into the books.
One route, built from first principles, that assembles a working MCP exchange:
flowchart LR
F["Fundamentals\nspecification, JSON-RPC, stateless core"] --> A["Architecture\ntopology, discovery, schemas"]
A --> I["Interactions\nprimitives, multi round-trip, errors, tasks"]
I --> S["Security\ntrust zones, OAuth, consent, audit"]
S --> U["Use cases\nroles, extensions, apps, registry"]
U --> C["Capstone\none 2026-07-28 exchange, end to end"]
Each lesson ships a runnable standard-library MCP mock, a test suite, a lesson
quiz, and a reusable artifact. Every lesson teaches the stateless 2026-07-28
revision as current, and scripts/check_mcpa_wire.py checks each lab's
transcript for that wire shape. The protocol facts, their primary sources, and
the source conflicts resolved while writing live in
research/mcp-2026-07-28-brief.md. The route
includes a 30-question diagnostic and three full-length original mocks, each
with a different emphasis, whose question mix follows the published blueprint
weights within practical rounding. They do not imitate or reproduce live exam
questions.
The MCPA blueprint has five domains:
| Domain | Weight |
|---|---|
| MCP Fundamentals | 16% |
| Architecture and Components | 14% |
| Interactions and Execution | 26% |
| Security and Governance | 24% |
| Use Cases and Ecosystem | 20% |
The tutor reads the track file for route order. This complete index also makes every lesson directly browsable from GitHub.
This is an independent community curriculum. It is not affiliated with, endorsed by, sponsored by, or authorized by the Agentic AI Foundation or the Linux Foundation. It does not contain live exam questions. The official exam page and current program policies always take precedence.