Internal Study Guide
2026 · v1.0
CCA-F
Claude Certified
Architect
Foundations
How our team prepares — and passes on the first attempt.
Certification
Anthropic · Pearson VUE
Audience
Solution architects & builders
How to use this guide
Read it once.
Then build.
This is a scenario exam. You pass by applying patterns, not by memorising bullet points. Pair every slide with hands-on practice.
What's inside
- The exam — what it is, logistics, how to register
- The 5 domains — weights + what to study in each
- The 6 scenarios — the contexts questions are built on
- Official resources — free Anthropic Academy courses
- A 3-week plan — plus a hands-on checklist
- Exam-day tactics — traps, timing, strategy
What is the CCA-F?
Anthropic's first
technical certification.
A proctored exam for solution architects who design and ship production systems with Claude — across the Claude API, the Claude Agent SDK, the Model Context Protocol (MCP) and Claude Code.
Announced March 12, 2026 as part of the Claude Partner Network. It is the Foundations tier — the entry point of a wider credential stack.
The program has three roles
- Practitioner
- Architect ← you are here
- Developer
Pearson VUE exam code: CCAR-F. Digital badge issued via Credly.
Who should sit it
Builders with real
Claude mileage.
There is no formal prerequisite, but the exam assumes hands-on experience. A candidate who already builds with Claude can be ready in 2–4 weeks.
You'll be tested on your ability to
- Choose the right architecture for a production use case
- Design tools and MCP integrations that scale
- Configure Claude Code for real engineering workflows
- Engineer prompts and reliable structured output
- Manage context, cost, latency and failure modes
Exam at a glance
The format, by the numbers.
~720 /1000
Pass mark (scaled)
1
Correct answer · 3 distractors
4 of 6
Scenarios drawn per exam
Verify before you book Question count, timing, pass mark and price are compiled from community & third-party prep sources — Anthropic has not published a full public exam guide. Partner Network members may sit it free. Always confirm current details in the Anthropic Partner Academy and your Pearson VUE portal.
How to register
Four steps to a booked seat.
1
Access the Partner Academy
Get into the Claude Partner Network → Anthropic Partner Academy, where prep and registration live.
2
Get Pearson credentials
Begin scheduling and receive a Pearson VUE username & password by email.
3
Schedule the exam
Book online-proctored or at a test centre. Reschedule / cancel up to 24h before.
4
Pass & claim the badge
Clear the pass mark; your digital badge is issued via Credly.
Retake policy Wait 14 days after attempt 1, 30 days after attempt 2, 90 days after attempt 3 — up to 4 attempts per rolling 12 months. (Source: Pearson VUE.)
The exam blueprint
Five domains.
Weighted by what
matters in production.
Agentic architecture and Claude Code carry the most weight — study depth accordingly.
D1
Agentic Architecture & Orchestration
27%
D3
Claude Code Configuration & Workflows
20%
D4
Prompt Engineering & Structured Output
20%
D2
Tool Design & MCP Integration
18%
D5
Context Management & Reliability
15%
Domain 1 · highest weight
Agentic Architecture
& Orchestration
27%
Know cold
- The augmented LLM building block: model + tools + retrieval + memory
- Workflows vs. agents — and choosing the simplest thing that works
- Workflow patterns: prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer
- The agent loop: act → observe tool results → iterate to a goal
- Multi-agent / subagent systems & the Claude Agent SDK
Study
- “Building effective agents” — Anthropic engineering
- Building with the Claude API (agentic sections)
- Introduction to Subagents
Trap Don't reach for a multi-agent system when a single workflow is more reliable and cheaper.
Domain 3
Claude Code
Configuration & Workflows
20%
Know cold
- CLAUDE.md project memory & instructions
- settings.json — permissions, env, hooks
- Slash commands & custom subagents
- Hooks for deterministic automation
- Wiring MCP servers into Claude Code
- Headless / CI mode; Agent Skills (SKILL.md)
Study
- Claude Code in Action
- Introduction to Agent Skills
- Claude Code docs — settings, hooks, subagents
Do Actually configure a repo: CLAUDE.md + a subagent + a hook + one skill. Muscle memory beats recall.
Domain 4
Prompt Engineering
& Structured Output
20%
Know cold
- Be clear & direct; give Claude a role (system prompt)
- XML tags to structure inputs and sections
- Multishot (few-shot) examples
- Chain-of-thought / “think” for hard reasoning
- Prefill the assistant turn to steer format
- Structured output via tool use → forced JSON schema
Study
- Prompt engineering guide — docs.anthropic.com
- Building with the Claude API (prompting + tools)
- Tool use /
tool_choice for guaranteed JSON
Key idea Reliable JSON comes from tools + a schema, not from asking nicely.
Domain 2
Tool Design
& MCP Integration
18%
Tool use — know cold
- A tool = name + description + input_schema (JSON Schema)
- Descriptions do the work — precise, minimal params
tool_choice, parallel tools, returning results & errors
MCP — know cold
- Open protocol connecting Claude to tools & data
- Servers expose Tools · Resources · Prompts
- Clients (Claude Code, Desktop); transports stdio / HTTP
- When to build a server vs. inline tools
Study Introduction to Model Context Protocol · Model Context Protocol: Advanced Topics · build one MCP server end-to-end.
Domain 5
Context Management
& Reliability
15%
Know cold
- Context windows & token budgets
- Prompt caching for cost & latency
- RAG / retrieval + citations to ground answers
- Reducing hallucination; allowing “I don't know”
- Evals & guardrails; retries, rate limits, streaming
Model selection
- Opus — deepest reasoning
- Sonnet — balanced default
- Haiku — fast & low-cost
Trap Reaching for the biggest model when a smaller one meets the SLA at a fraction of the cost.
The 6 scenario contexts
Questions live inside
real build scenarios.
Each exam draws 4 of these 6. Practise applying the domains to each context.
01
Customer Support Resolution Agent
02
Code Generation with Claude Code
03
Multi-Agent Research System
04
Developer Productivity Tools
06
Structured Data Extraction
Official study resources
Free, self-paced,
and exam-aligned.
Anthropic Academy (Skilljar) maps directly to the domains. Start with the flagship API course, then MCP, then Claude Code.
Registration & exam-prep material live in the Anthropic Partner Academy; the courses below are public on anthropic.skilljar.com.
Anthropic Academy — course path
- Building with the Claude API ~8.1 h · flagship
- Introduction to Model Context Protocol
- Model Context Protocol: Advanced Topics
- Claude Code in Action
- Introduction to Agent Skills
Plus: docs.anthropic.com · “Building effective agents” · Prompt engineering guide.
The stack you're certified on
Four pillars. Know how they fit.
Claude API
Messages API, system prompts, tool use, streaming, prompt caching. The foundation everything else builds on.
Claude Agent SDK
Build agents that plan, call tools in a loop, and coordinate subagents toward a goal.
MCP
Open protocol to connect Claude to external tools, data and prompts through servers & clients.
Claude Code
Agentic coding in the terminal — configured with memory, permissions, subagents, hooks and skills.
A 3-week plan
Three weeks, if you're
already hands-on.
Week 1 · Foundations
API & prompting
Building with the Claude API + prompt engineering. Build a tool-using app that returns validated JSON. Covers D4.
Week 2 · Integration
MCP & Claude Code
Intro + Advanced MCP, Claude Code in Action, Agent Skills. Ship an MCP server; configure a real repo. Covers D2, D3.
Week 3 · Agents + review
Architecture & reliability
Building effective agents, multi-agent, context & cost. Drill the 6 scenarios + mock questions. Covers D1, D5.
Hands-on checklist
You can't read
your way to a pass.
Before you book, make sure you have personally built each of these at least once.
- A tool-use loop: model calls a tool, reads the result, continues
- An MCP server exposing a tool and a resource
- A Claude Code repo: CLAUDE.md + subagent + hook + skill
- A structured-extraction pipeline with a strict schema
- A small multi-agent / orchestrator-worker system
- A run with prompt caching + a basic eval
Exam-day tactics
Scenario-based, single answer.
Play it accordingly.
Before answering
- Read the scenario's constraints first — scale, latency, cost, safety
- Note what “production” implies for this question
Choosing
- Eliminate distractors — 1 right, 3 wrong
- Prefer the simplest robust pattern that meets the constraint
Timing
- ~2 min / question; flag & return
- No external tools — no Claude, no docs
Mindset The “best” answer is usually the one an experienced architect would ship — reliable, cost-aware, and no more complex than it needs to be.
Common pitfalls & distractor traps
- Over-engineering — an agent where a workflow is enough
- Tool use vs. MCP — confusing the API feature with the protocol
- Ignoring cost — forgetting prompt caching & model tiering
- No guardrails — skipping evals, validation, “I don't know”
- Wrong model — Opus where Haiku meets the SLA
- Format by hope — asking for JSON instead of enforcing a schema
One-page cheat sheet
Domain weights
D1 Agentic 27 · D3 Claude Code 20 · D4 Prompting 20 · D2 Tools/MCP 18 · D5 Context 15
Workflow patterns
Prompt chaining · routing · parallelization · orchestrator-workers · evaluator-optimizer
MCP primitives
Server: Tools · Resources · Prompts. Client: sampling, roots. Transports: stdio / HTTP.
Claude Code surfaces
CLAUDE.md · settings.json · slash commands · subagents · hooks · MCP · skills · headless
Structured output
Tool use + JSON Schema + tool_choice. Prefill to steer. Validate the result.
Model tiers
Opus = reasoning · Sonnet = balanced · Haiku = fast/cheap. Match to the SLA.
Sources & disclaimer
Official
- anthropic.com — Claude Partner Network announcement (Mar 12, 2026)
- pearsonvue.com/us/en/anthropic — registration & retake policy
- anthropic.skilljar.com — Anthropic Academy courses
- docs.anthropic.com — API, tools, MCP, prompt engineering
Community (exam mechanics)
- Third-party prep guides — domain weights, 6 scenarios, question count
- Used only where multiple independent sources agree
Disclaimer Anthropic has not published a full public exam guide. Treat all logistics as indicative and confirm current details in the Anthropic Partner Academy and Pearson VUE before booking.
Team next steps
Book it.
Build it.
Pass first try.
Pick your exam date, work the 3-week plan, and ship every item on the hands-on checklist. Reading is the warm-up — building is the exam.
CCA-F · Team Study Guide
Questions → #eng-certs