CCA-F Exam Study Guide

How the Exam Is Structured

The shared CCA-F outline is organized into five domains and approximately 50 multiple-choice questions. The ExamPal practice bank includes 492 premium questions and 40 free questions mapped across the same domains.

19% of exam

Domain 2: Agentic Architecture & Orchestration

Covers agent design patterns, orchestration strategies, asynchronous execution, handoff workflows, and core Agent SDK concepts. This domain focuses on choosing the right agent structure, coordinating work across agents, and managing lifecycle and context flow in agentic systems. Candidates should be ready to connect these concepts to realistic architecture and workflow decisions, especially where configuration scope, tool boundaries, context management, and operational reliability affect production use.

19% of exam

Domain 4: Tool Design & MCP Integration

Covers tool interface design, MCP server integration, structured error handling, and tool selection behavior. This domain focuses on building reliable tools and integrating them safely through MCP with appropriate trust, authentication, and error semantics. Candidates should be ready to connect these concepts to realistic architecture and workflow decisions, especially where configuration scope, tool boundaries, context management, and operational reliability affect production use.

Key Terms to Know

These terms are loaded from the CCA-F terminology release pack and appear across the question explanations.

API key

A credential used to authenticate access to an API or service.

CLAUDE.md

A persistent instruction file that provides guidance to Claude Code sessions but does not enforce runtime actions.

context window

The bounded amount of information a model can access and process in a single interaction.

conversation history

The accumulated prior messages and interactions available to an agent as context.

coordinator-subagent pattern

An orchestration design where a coordinator delegates subtasks to subagents and combines their results.

custom slash command

A reusable command invoked with a slash prefix to package multi-step workflows into one action.

errorCategory

A structured field classifying the type of error returned by a tool or system.

ETL

Extract, transform, load; a sequential data-processing pipeline where each stage depends on the previous one.

hook

An execution-time interception mechanism that can enforce, modify, or block actions.

intermediate context

Temporary working information generated during subtask processing before final aggregation.

isError

A response field indicating whether a tool call resulted in an error.

latency

The time delay between initiating a request and receiving its result.

lifecycle events

Named stages in Claude Code execution, such as session start, prompt submission, tool use, and stop.

MCP tool

A tool integrated through the Model Context Protocol that returns structured responses to agent requests.

model calls

Requests made to an AI model for inference or generation.

overhead

Additional computational or operational cost introduced by a design choice without proportional benefit.

parallelism

The ability to perform multiple tasks concurrently to reduce overall latency.

refresh policies

Rules that determine when stored data should be updated to maintain accuracy.