Claude Certified Architect - Foundations (CCAR-F) Flashcards¶

120 minutes

60 multiple-choice and multiple-response, scenario-based

720 / 1000

$125 USD

12 months, free non-proctored on-time renewal

Pearson VUE (online proctored or test center)

Claude decides which tools to call and in what order

Multiple rounds of tool calls and reasoning

Claude works toward completing a defined objective

The path is not predetermined; it depends on intermediate results

Production rate management

~/.claude/settings.json - Personal settings

.claude/settings.json - Project-specific settings (checked into repo)

Managed centrally by administrators

Tools that can run without asking for permission

Tools that are blocked entirely

Unspecified tools prompt for approval

Runs before a tool is executed. Can block the tool execution.

Runs after a tool completes. Can process or validate results.

Runs when Claude Code sends a notification

Run linter before file writes to ensure code quality

Check that proposed changes meet project standards

Scan for secrets or vulnerabilities before committing

Auto-format files after Claude writes them

Run relevant tests after code changes

Alert the team when significant changes are made

For local MCP servers. Claude Code spawns the process and communicates via stdin/stdout.

For remote MCP servers. Claude Code connects via HTTP.

Claude clearly distinguishes between instructions, context, and data

You can ask Claude to wrap its output in XML tags for easy extraction

Complex prompts become readable and maintainable

You can refer to tagged sections by name (e.g., "use the data in ")

Task is straightforward and Claude handles it well without examples

Format or behavior needs demonstration; common choice

Highly specific or nuanced task requiring precise pattern matching

Controls randomness in output (0 to 1)

Alternative to temperature, controls diversity

Limits the number of tokens considered at each step

One protocol for all tool integrations

Build a server once, use it with any MCP client

Clients can discover available tools, resources, and prompts

Defined authorization and capability negotiation

The AI application (e.g., Claude Desktop, Claude Code, your custom app)

The MCP client within the host that manages server connections

A service that exposes tools, resources, and prompts via MCP

Use streamable HTTP

Use streamable HTTP

Read-only data access. The model requests data by URI.

Executable functions with parameters. The model calls a function.

@modelcontextprotocol/sdk

Thinking tokens add to the total cost of each request

Extended thinking takes additional time before the response

Generally improves quality for complex tasks

Thinking content is delivered as a single event, then response streams normally

Start with 1-2 seconds

Double the delay on each retry

Add random variation to prevent thundering herd

Cap at 30-60 seconds

Typically 3-5 attempts

Maximum number of API requests per minute

Maximum input + output tokens per minute

Maximum tokens per day (some tiers)

Free official courses

Primary reference