Install Claude Code: Complete Developer Setup Guide

If you need to install Claude Code and get it running on a real project, this guide covers every step: prerequisites, the install command, first-run configuration, and the project-level settings that separate a working setup from a production-ready one. We also map each setup decision to the Claude Certified Developer, Foundations (CCDV-F) exam, so the time you spend here doubles as exam prep.

What is Claude Code and why does it matter for the CCDV-F exam?

Claude Code is Anthropic's agentic coding tool. It runs in your terminal, reads your codebase, calls tools, edits files, and executes shell commands autonomously inside a session. For developers preparing for the CCDV-F exam, it is not optional background knowledge: Domain 3 (Claude Code) and Domain 2 (Applications and Integration, weighted at 33.1%) both test practical judgment about how Claude Code behaves in real workflows.

The CCDV-F exam launched on 12 March 2026 alongside the full Claude Partner Network certification programme. It costs $125 USD per attempt, runs 53 items in 120 minutes, and requires a scaled score of 720 out of 1000 to pass. Understanding the tool at the command-line level is the foundation for every higher-order question the exam asks.

What are the prerequisites before you install Claude Code?

Before you run the install command, confirm three things:

1. Node.js 18 or later. Claude Code is distributed as an npm package. Run node --version to check. If you are below 18, update via nodejs.org or your version manager of choice.
2. npm 9 or later (ships with Node 18+). Run npm --version.
3. An Anthropic API key. Claude Code authenticates against the Anthropic API. Generate a key at console.anthropic.com. Store it somewhere safe before you proceed.

You do not need a separate Claude.ai subscription to use Claude Code via the API, though some teams route usage through Claude.ai Pro or Team plans. The CCDV-F exam tests API-level integration, so working directly with an API key is the right mental model.

How do you install Claude Code?

The install is a single npm command:

npm install -g @anthropic-ai/claude-code

The -g flag installs the package globally so the claude binary is available in any directory. After the install completes, verify it:

claude --version

You should see a version string. If the command is not found, your global npm bin directory is likely not on your PATH. Fix it by running:

npm config get prefix

Then add <prefix>/bin to your shell's PATH in .bashrc, .zshrc, or the equivalent.

What if you are on a managed or restricted machine?

Some corporate environments block global npm installs. In that case, use npx to run Claude Code without a global install:

npx @anthropic-ai/claude-code

This is slower on first run because it downloads the package each time, but it works without elevated permissions. For CI/CD pipelines, npx is often the right choice anyway because it pins the version per invocation.

How do you configure Claude Code after installation?

Setting your API key

On first run, Claude Code will prompt for your API key if it is not already set. You can also set it as an environment variable, which is the recommended approach for any non-interactive context:

export ANTHROPIC_API_KEY="sk-ant-..."

Add that line to your shell profile to persist it across sessions. For CI/CD, inject it as a secret rather than hardcoding it in any file.

Running Claude Code for the first time

Navigate to a project directory and run:

claude

Claude Code opens an interactive session. It reads the current directory, and if a CLAUDE.md file exists at the project root, it loads that as its primary instruction context. If no CLAUDE.md exists, Claude Code operates with its defaults.

What is CLAUDE.md and why does it matter?

CLAUDE.md is a markdown file you place at the root of your repository (or in subdirectories for path-scoped rules). Claude Code reads it at the start of every session and treats its contents as persistent instructions. Think of it as a system prompt that lives in version control.

A minimal CLAUDE.md for a Python project might look like this:

# Project: Payments Service

## Stack
- Python 3.12, FastAPI, PostgreSQL
- Tests: pytest with coverage enforced at 90%

## Conventions
- All database queries go through the repository layer; never query directly from route handlers.
- Use `ruff` for linting before committing.

## Out of scope
- Do not modify migrations without explicit instruction.

The three-level configuration hierarchy that Claude Code uses (global user settings, project-level CLAUDE.md, and path-scoped rules) is a testable concept on the CCDV-F exam. Understanding which level takes precedence, and when to use each, is the kind of practical judgment the exam rewards.

The version control implications of committing CLAUDE.md are also worth understanding: a CLAUDE.md checked into a shared repository becomes a team-wide instruction set, which is powerful but requires the same review discipline as any other configuration file.

How do you configure Claude Code for a team environment?

Individual developers can keep personal preferences in a global config file (~/.claude/CLAUDE.md), while shared project conventions live in the repository-level CLAUDE.md. This separation means a developer can have personal shortcuts without polluting the shared config.

For teams, the recommended pattern is:

Path-scoped configs let you apply different rules to different parts of a monorepo. A CLAUDE.md inside src/payments/ applies only when Claude Code is working within that directory tree, which is useful when different sub-teams have different conventions or when certain directories contain sensitive logic that should not be autonomously modified.

What MCP servers can you connect after installation?

Claude Code supports the Model Context Protocol (MCP), which lets you connect external tools and data sources to a Claude Code session. After installation, you can configure MCP servers in your project or user-level config.

A basic MCP server entry in a Claude Code config looks like this:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/project"],
      "env": {
        "NODE_ENV": "production"
      }
    }
  }
}

The environment variable expansion in MCP config is a detail the CCDV-F exam tests directly: knowing that ${VAR} syntax expands at config-load time, not at runtime, affects how you structure secrets and paths. Domain 8 (Tools and MCPs) carries 10.6% of the exam weight, so MCP configuration is worth understanding thoroughly.

How does Claude Code handle tool use in a session?

Once installed and running, Claude Code has access to a set of built-in tools: file reading and writing, shell command execution, search (grep and glob), and the Edit tool for targeted in-place modifications. Understanding when Claude Code reaches for each tool is directly testable.

The distinction between grep and glob is a common exam scenario: grep searches file contents, glob matches file paths by pattern. Choosing the wrong one wastes tokens and produces incorrect results. Similarly, the Edit tool and read/write fallback pattern matters when you need surgical edits versus full file rewrites.

For the CCDV-F exam, the built-in tool sequence for common tasks is worth memorising as a pattern: explore with glob, read with the Read tool, edit with the Edit tool, verify with shell commands. Deviating from this sequence without reason is a signal of poor tool judgment.

What does the CCDV-F exam actually test about Claude Code?

Domain 3 (Claude Code) carries 3.1% of the exam weight, which is the smallest domain by weight. Do not let that number mislead you: Claude Code concepts bleed into Domain 2 (Applications and Integration, 33.1%) and Domain 8 (Tools and MCPs, 10.6%), because real integration work involves Claude Code sessions, MCP connections, and tool-use decisions.

Here is the full domain weight breakdown for context:

The exam does not draw from a scenario bank the way the CCAR-F Architect exam does. CCDV-F items are written directly against the skills in each domain, which means breadth of coverage matters more than deep preparation for a handful of scenarios.

How do you use Claude Code non-interactively in CI/CD?

For automated pipelines, Claude Code supports a non-interactive mode via the -p flag (print mode). This lets you pass a prompt directly and receive output without an interactive session:

claude -p "Review the diff in the last commit and list any security concerns" --output-format json

The -p flag is essential for CI/CD integration. Without it, Claude Code waits for interactive input, which will hang a pipeline. Pairing it with --output-format json gives you structured output you can parse downstream.

For the exam, understanding the distinction between interactive sessions (where Claude Code can ask clarifying questions and iterate) and non-interactive invocations (where it must complete the task in a single pass) maps directly to Domain 2 questions about application integration patterns.

What should you do after installation to prepare for the CCDV-F exam?

Installation is the starting point, not the destination. After you have Claude Code running, the highest-leverage preparation steps are:

1. Build a CLAUDE.md for a real project and observe how it changes Claude Code's behaviour across sessions.
2. Connect at least one MCP server and trace how tool selection changes when more tools are available.
3. Run Claude Code in non-interactive mode against a test suite and inspect the structured output.
4. Read the prompt engineering and structured output concepts that underpin Domain 6, since prompt quality directly affects Claude Code's tool-use decisions.
5. Study agentic architecture patterns to understand how Claude Code sessions fit into larger multi-agent systems, which Domain 1 tests at 14.7% weight.

The CCDV-F exam costs $125 USD and runs 53 items in 120 minutes. At roughly 2.3 minutes per item, there is no time to reconstruct knowledge from first principles. The goal is to have internalised the decision rules well enough that each scenario resolves quickly.

AI Skill Certs is an independent prep platform and is not affiliated with or endorsed by Anthropic. Our CCDV-F prep content is in development; our live CCAR-F Architect Foundations prep is available now if you are also pursuing the Architect track.