Skip to content
Study sections
1. Exam Overview

Section 1 — Exam Overview, Structure & Scenarios

What this section covers

A grounding in why the Claude Certified Architect – Foundations (CCA-F) exists, what it tests, how it is structured (domains, scoring, scenarios), and how to plan a study path that maps cleanly to the published domain weights.

Source material (from official guide)

Purpose. The CCA-F validates that practitioners can make informed tradeoff decisions when implementing real-world solutions with Claude. The exam covers four core technologies: Claude Code, the Claude Agent SDK, the Claude API, and the Model Context Protocol (MCP).

Target candidate. A solution architect with roughly 6+ months of hands-on Claude experience who has personally:

  • Built agentic apps with the Claude Agent SDK (orchestration, subagents, tool integration, lifecycle hooks).
  • Configured Claude Code for teams via CLAUDE.md, Agent Skills, MCP servers, and plan mode.
  • Designed MCP tool and resource interfaces against real backends.
  • Engineered prompts that produce reliable structured output (JSON schemas, few-shot, extraction patterns).
  • Managed context windows across long documents and multi-agent handoffs.
  • Integrated Claude into CI/CD for code review, test generation, and PR feedback.
  • Made escalation and reliability decisions including human-in-the-loop and self-evaluation.

Question format. Multiple choice, one correct answer plus three distractors. There is no penalty for guessing; unanswered questions are scored as incorrect.

Scoring. Pass/fail with a scaled score of 100–1,000. The minimum passing score is 720. Scaled scoring equates results across exam forms with slightly different difficulty.

Domains and weightings.

#DomainWeight
1Agentic Architecture & Orchestration27%
2Tool Design & MCP Integration18%
3Claude Code Configuration & Workflows20%
4Prompt Engineering & Structured Output20%
5Context Management & Reliability15%

Scenarios. Each exam draws 4 of 6 production scenarios at random. The full set:

  1. Customer Support Resolution Agent — Agent SDK agent over MCP tools (get_customer, lookup_order, process_refund, escalate_to_human); target 80%+ first-contact resolution with sound escalation.
  2. Code Generation with Claude Code — CLAUDE.md, custom slash commands, plan mode vs. direct execution.
  3. Multi-Agent Research System — Coordinator delegating to web-search, document-analysis, synthesis, and report-generation subagents.
  4. Developer Productivity with Claude — Agent SDK over built-in tools (Read, Write, Bash, Grep, Glob) plus MCP servers for codebase exploration and boilerplate generation.
  5. Claude Code for Continuous Integration — CI/CD-integrated code review, test generation, and PR feedback; minimize false positives.
  6. Structured Data Extraction — JSON-schema-validated extraction from unstructured documents with graceful edge-case handling.

Out of scope. Fine-tuning, billing/auth/quotas, vision and computer use, streaming API internals, embedding/vector DBs, RLHF/Constitutional AI, prompt-caching internals, tokenization, specific cloud-provider configs, and similar deployment plumbing topics.

Enriched context

Certification ecosystem

Anthropic publicly launched the CCA-F on March 12, 2026 as its first official technical certification. The format that has been reported by multiple secondary sources is 60 multiple-choice questions, 120 minutes, proctored, closed-book, with results delivered within ~2 business days including section breakdowns. Cost is $99 per attempt, waived for the first 5,000 attempts allocated to the Claude Partner Network.

Access today is organization-mediated through the Claude Partner Network rather than fully self-serve: candidates either work at a partner org or align with one, then request an exam slot on anthropic.skilljar.com. The supporting curriculum — Anthropic Academy — is free and lives on Skilljar; it includes “Claude 101,” “Building with the Claude API,” MCP introductory and advanced courses, “Claude Code in Action,” “Introduction to agent skills,” “Introduction to subagents,” and additional cloud-deployment courses.

Anthropic has signaled that additional tiers for sellers, developers, and advanced architects are coming later in 2026. Treat the CCA-F as the entry level of a multi-tier credential stack rather than a one-off badge.

Domain weighting strategy

The five domains are tightly clustered (15%–27%), so no domain can be skipped, but Domain 1 (Agentic Architecture & Orchestration) is the single largest lever. A pragmatic study budget based on weights:

  • ~27% of study time on Domain 1 — agent loops, stop_reason handling, subagent delegation, lifecycle hooks. This domain also shows up inside Scenarios 1, 3, and 4, so the leverage is higher than the raw weight suggests.
  • ~20% each on Domains 3 and 4 — Claude Code configuration (CLAUDE.md hierarchy, plan mode, slash commands, skills) and prompt/structured-output engineering (JSON schemas, few-shot, retry loops).
  • ~18% on Domain 2 — MCP tool descriptions that disambiguate similar tools, structured error responses with retryable flags, and resource design.
  • ~15% on Domain 5 — context-window management, scratchpads, escalation rules, confidence-based human-in-the-loop routing.

A useful heuristic: roughly two-thirds of the exam turns on agent-loop design, Claude Code configuration, and prompt/structured-output decisions. Spend study time there before tuning the long tail.

The six scenarios — what to expect

  • Scenario 1 — Customer Support Resolution Agent. Expect questions about when the agent escalates (policy gap, explicit customer request, inability to make progress) versus resolves autonomously, and how process_refund should expose retryable/non-retryable errors so the agent can recover without looping.
  • Scenario 2 — Code Generation with Claude Code. Expect plan-mode-vs-direct-execution tradeoffs, where path-specific rules live (.claude/rules/), and how custom slash commands and skills are scoped (context: fork, allowed-tools).
  • Scenario 3 — Multi-Agent Research System. Expect coordinator/subagent boundaries, how to pass cited context between agents without blowing the window, and where to enforce evaluator/critic loops to keep citations accurate.
  • Scenario 4 — Developer Productivity with Claude. Expect tool-selection questions across built-ins (Read, Write, Bash, Grep, Glob) plus MCP add-ons, and judgments about when to keep work in the parent agent vs. fork a subagent.
  • Scenario 5 — Claude Code for CI. Expect prompt-engineering questions for actionable review comments, explicit review criteria to suppress false positives, and multi-pass review patterns for large diffs.
  • Scenario 6 — Structured Data Extraction. Expect schema design with nullable/optional fields, validation-retry loops on tool_use, batch processing tradeoffs, and graceful degradation when a document is partially malformed.

Comparable certifications

CertificationVendorFocusNotes
Claude Certified Architect – FoundationsAnthropicProduction Claude architecture: agents, MCP, Claude Code, structured outputScenario-heavy, judgment-driven; partner-mediated access; $99
AWS Certified AI Practitioner (AIF-C01)AWSFoundational AI/ML literacy + Bedrock concepts85 questions / 120 min; passing 700/1000; broader and more conceptual, less hands-on
Google Cloud Generative AI LeaderGoogleBusiness strategy for GenAI on Vertex AI / GeminiLeadership-oriented, not implementation-focused

The clearest contrast: AWS AI Practitioner and GCP Generative AI Leader test breadth and literacy across managed services; CCA-F tests depth and judgment inside a single vendor’s agentic stack. Multi-cloud architects often stack them rather than choose.

Key terms glossary

  • Agent SDK — Anthropic’s library for building agentic loops with tool calling, subagents, and lifecycle hooks.
  • Claude Code — Anthropic’s terminal-native coding agent, configured via CLAUDE.md, slash commands, Agent Skills, and MCP servers.
  • MCP (Model Context Protocol) — Open protocol for exposing tools, resources, and prompts from backend systems to Claude.
  • Plan mode — Claude Code mode that produces a plan before any side-effecting action runs.
  • CLAUDE.md — Hierarchical project/user-level configuration file consumed automatically by Claude Code.
  • Scaled score — A 100–1,000 normalized score that equates across exam forms; the cut score is 720.
  • Scenario — A realistic production context that frames a batch of related questions; 4 of 6 appear per exam.
  • Distractor — A plausible-looking wrong answer that catches candidates with incomplete experience.

Recommended preparation path

  1. Skim the 12 study sections in order, paying extra attention to the task statements quoted at the top of each section — they are the per-domain learning objectives drawn straight from the official exam guide.
  2. Complete the free Anthropic Academy courses on Skilljar in this order: Claude 101 → Building with the Claude API → Introduction to MCP → MCP Advanced Topics → Claude Code in Action → Introduction to agent skills → Introduction to subagents.
  3. Build one Agent SDK project from scratch that exercises a full agentic loop: tool calling, stop_reason handling, subagent delegation, lifecycle hooks, and error/retry policy.
  4. Configure Claude Code for a real repo: layered CLAUDE.md, .claude/rules/ path-specific rules, at least one custom slash command, one skill with context: fork and allowed-tools, and one MCP server integration.
  5. Design and stress-test MCP tools: write tool descriptions that disambiguate near-duplicates, return structured errors with retryable flags, and verify tool selection on ambiguous prompts.
  6. Ship a structured extraction pipeline with JSON schemas, a validation-retry loop, optional/nullable fields, and the Message Batches API for throughput.
  7. Drill on the six scenarios by writing a one-page architecture sketch for each — primary domains, tool inventory, escalation policy, context strategy, and reliability patterns.
  8. Take the official practice exam last, review every explanation (including the ones you got right), and book the proctored sitting only once you are scoring comfortably above 720 on simulated material.

References

Last updated on
2. Agentic Loops