By Connectry Labs

Pass the Architect Cert

390 scenario-based questions. Spaced repetition. Guided capstone build. Ask Claude, study smart, pass the Claude Certified Architect exam. Free. Local. No fluff.

390 Questions
30 Handouts
6 Projects
5 Exam Domains
Get Started View on GitHub

Study Through Conversation

Ask Claude anything. Architect Cert handles the rest.

You
I want to start studying for the Architect exam
C
I'll run a 15-question diagnostic to baseline your knowledge across all 5 domains.

This takes about 10 minutes and will tell us where to focus first. Ready?

Start assessment Skip to practice
You
Give me a hard question on agentic architecture
C
Domain 1 — Task 1.2 (Hard)

A coordinator agent delegates to 4 subagents in parallel. Two return conflicting results on the same data point. What is the architecturally correct resolution strategy?

A) Use the first result received — fastest wins
B) Retry both subagents with a different prompt
C) Escalate to a dedicated arbitration agent with both outputs
D) Default to the subagent with higher token usage as a quality proxy
You
C
C
✓ Correct. Dedicated arbitration keeps conflict resolution as an explicit architectural concern — it preserves audit trails and separates coordination logic from execution logic.

Why the others are wrong:
• A: Ignores correctness entirely — speed is not a quality signal
• B: Reprompting non-deterministic agents doesn't guarantee convergence
• D: Token usage is not correlated with output quality

Next question Show handout See reference project

Everything You Need to Pass

A complete study system — not a question dump

390 Scenario Questions

Scenario-based multiple choice across all 5 domains and 30 task statements. Easy, medium, and hard — balanced answer distributions. No trick questions.

SM-2 Spaced Repetition

The SM-2 algorithm schedules reviews at optimal intervals. Wrong answers resurface sooner. Mastered topics fade out. You always work on what matters for exam readiness.

Guided Capstone Build

Shape your own project, then build it step-by-step across 18 guided steps. Each step: quiz on the task statements you're about to implement, then Claude generates the code. Learn by building.

30 Concept Handouts

One per task statement — concept explanation, TypeScript code example, common mistakes, and documentation references. Available in Claude or as branded PDFs for offline study.

6 Reference Projects

Runnable TypeScript codebases — one per domain, plus a capstone tying all 5 together. Every file maps to specific task statements. See concepts in real production-grade code.

Practice Exams

Full 60-question exams weighted exactly like the real certification. Scored 0–1000, passing at 720. All attempts saved with per-domain breakdowns and improvement trends.

Deterministic Grading. Zero Sycophancy.

gradeAnswer() is a pure function. No LLM judges your answer. Right is right, wrong is wrong. Claude won't say "you were on the right track" — wrong means wrong, with a specific explanation of why. Anti-sycophancy rules are enforced at the protocol level.

How It Works

Install once. Study through Claude. Everything stays local.

1

Install

One command. No accounts, no keys, no config.

npm install -g connectry-architect-mcp
2

Connect

Add to Claude Desktop or Claude Code config

"command": "connectry-architect-mcp"
3

Study

Just ask Claude

"Start an assessment"
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "connectry-architect": {
      "command": "connectry-architect-mcp"
    }
  }
}
%APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "connectry-architect": {
      "command": "connectry-architect-mcp"
    }
  }
}
.mcp.json (project) or ~/.claude.json (global)
{
  "mcpServers": {
    "connectry-architect": {
      "command": "connectry-architect-mcp"
    }
  }
}

5 Exam Domains. 30 Task Statements.

Complete coverage of the Claude Certified Architect — Foundations exam

D1 27%

Agentic Architecture & Orchestration

7 tasks · 91 questions

  • Agentic loops for autonomous task execution
  • Multi-agent coordinator-subagent patterns
  • Session state, resumption, and forking
  • Task decomposition for complex workflows
D2 18%

Tool Design & MCP Integration

5 tasks · 65 questions

  • Effective tool interfaces and boundaries
  • Structured error responses for MCP tools
  • Tool distribution across agents
  • Built-in tool selection and application
D3 20%

Claude Code Configuration & Workflows

6 tasks · 78 questions

  • CLAUDE.md hierarchy and scoping
  • Custom slash commands and skills
  • CI/CD pipeline integration
  • Plan mode vs direct execution
D4 20%

Prompt Engineering & Structured Output

6 tasks · 78 questions

  • Explicit criteria for precision
  • Few-shot prompting for consistency
  • Structured output with tool use and JSON schemas
  • Validation, retry, and feedback loops
D5 15%

Context Management & Reliability

6 tasks · 78 questions

  • Context preservation in long sessions
  • Escalation and ambiguity resolution
  • Error propagation in multi-agent systems
  • Confidence calibration and provenance

Multiple Ways to Study

Adaptive paths based on how you learn best

01

Adaptive Practice

Spaced repetition with priority-based question selection. Overdue reviews first, weak areas second, new material third. Always working on what moves the needle.

"Give me a practice question"
02

Concept-First Learning

Read the handout before tackling questions on any topic. Concept, code example, common mistakes, documentation references — then 13 questions to test your understanding.

"Teach me about task 1.5"
03

Guided Capstone Build

Shape your own project idea. Claude validates 30-task-statement coverage, then builds it file-by-file across 18 steps — quiz, build, walkthrough. Learn by doing.

"Start a guided capstone build"
04

Practice Exams

Full 60-question weighted exams. Scored 0–1000, passing at 720 — exactly like the real thing. Per-domain breakdowns track improvement trends across attempts.

"I want to take a practice exam"

Everything Stays on Your Machine

No cloud. No accounts. No telemetry.

Progress stored locally in SQLite at ~/.connectry-architect/progress.db
No internet connection required after install
Zero telemetry — nothing is tracked or sent anywhere
Open source under MIT — inspect every line
Works with Claude Desktop, Claude Code, or any MCP client

Start Studying. Pass the Exam.

Install Architect Cert and open Claude. That's it.

npm install -g connectry-architect-mcp
View Source npm Package Connectry Labs

Frequently Asked Questions

Everything you need to know about Architect Cert

What is Architect Cert?

Architect Cert is a free, open-source MCP server that turns Claude into your personal certification tutor for the Claude Certified Architect — Foundations exam. It ships with 390 scenario-based questions, spaced repetition, a guided capstone build, 30 concept handouts, 6 reference projects, and practice exams.

Is Architect Cert free?

Yes, completely free. Architect Cert is open source under the MIT license. Install from npm, use it however you want. No payment required, no premium tier, no restrictions.

What MCP clients does it work with?

Architect Cert works with Claude Desktop (macOS and Windows), Claude Code (VS Code, Cursor, Terminal), and any other client that supports the Model Context Protocol over stdio. Configure it with one JSON snippet.

How is grading different from just asking Claude?

Grading is deterministic — a pure TypeScript function checks your answer against a verified key. No LLM is involved in judging correctness. Wrong answers are wrong with a specific explanation. Claude won't say "you were close" — that's the point.

Where is my progress stored?

Locally on your machine at ~/.connectry-architect/progress.db (SQLite, WAL mode). No cloud, no accounts, no telemetry. Your data never leaves your machine.

What's included in the guided capstone build?

You describe a project idea, Claude validates it covers all 30 task statements, then builds it file-by-file across 18 steps. Each step: quiz on the relevant task statements, Claude generates the code, then a walkthrough explains every section. After step 18 you get a full coverage map.

Copied to clipboard!