Getting Started with Claude Code via the Anthropic API (Enterprise Setup)

Claude Code is Anthropic’s terminal-based coding agent. Most individuals log in with a Claude.ai subscription (Pro/Max/Team), but if your org runs on the Anthropic API (Console/Enterprise pay-as-you-go), you authenticate with an API key instead — which gives you per-key budgets, dedicated rate limits, and full cost visibility in the Console dashboard.

This tutorial covers install → auth → your first session → beginner-friendly things to try.

1. Install Claude Code

You need Node.js installed first, then:

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

Verify it installed:

claude --version

2. Authenticate with an API key (enterprise/org path)

For an org using the Anthropic API rather than individual subscriptions, an admin creates the key in Claude Console and shares it with the team (ideally through a secrets manager, not Slack/email).

Set it as an environment variable:

export ANTHROPIC_API_KEY="your-api-key-here"

# Make it permanent by adding to your shell profile
echo 'export ANTHROPIC_API_KEY="your-api-key-here"' >> ~/.bashrc
source ~/.bashrc

Notes for enterprise/team use:

• Anthropic recommends a shared org-issued key per team/project rather than every developer minting their own — it keeps billing, revocation, and usage monitoring centralized.
• If your org routes through Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry instead of a direct API key, that’s set via CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, or CLAUDE_CODE_USE_FOUNDRY env vars — ask your platform team which applies.
• If you ever see auth errors even though the key looks right, run unset ANTHROPIC_API_KEY to fall back to another credential, and use /status inside Claude Code to check which auth method is currently active.

If your company routes traffic through its own gateway URL

Many enterprises don’t let Claude Code hit api.anthropic.com directly — instead they run an internal gateway/proxy (for audit logging, budget enforcement, or multi-provider routing) that speaks the same Anthropic API shape. In that case, two variables matter, not just one:

export ANTHROPIC_BASE_URL="https://your-company-gateway.example.com"
export ANTHROPIC_AUTH_TOKEN="your-gateway-token"   # if the gateway issues its own token
# OR keep using ANTHROPIC_API_KEY if the gateway accepts the standard x-api-key header

• ANTHROPIC_BASE_URL only changes where requests go — it doesn’t change how you authenticate.
• Check with your platform team whether their gateway expects x-api-key (→ use ANTHROPIC_API_KEY) or a bearer token (→ use ANTHROPIC_AUTH_TOKEN). This is the single most common gateway misconfiguration.
• For it to persist across every session (not just your current shell), put it in ~/.claude/settings.json under the env block: { "env": { "ANTHROPIC_BASE_URL": "https://your-company-gateway.example.com", "ANTHROPIC_AUTH_TOKEN": "your-gateway-token" }}
• The value is read once, at process start — changing it while claude is running does nothing. Restart the session after any change.
• If the gateway is a non-Anthropic host, Claude Code disables MCP tool search by default (it relies on the gateway forwarding certain internal blocks). If your gateway forwards them correctly, your platform team can re-enable it with ENABLE_TOOL_SEARCH=true.
• Quick sanity check: echo $ANTHROPIC_BASE_URL to confirm it’s set, and /status inside Claude Code to confirm it’s the active endpoint.

3. Start your first session

Navigate to a project folder and run:

cd my-project
claude

This drops you into an interactive session. Just type in plain English — no special syntax needed.

4. What beginners can actually do with Claude Code

Here’s a practical, low-risk on-ramp — roughly in the order most people try things:

a) Ask questions about the codebase (read-only, zero risk)

> What does this project do?
> Where is the authentication logic handled?
> Explain how the payment flow works

b) Make a small, contained code change

> Add a loading spinner to the login button
> Fix the bug where dates display in the wrong timezone

Claude Code will show you a diff before applying changes — review it, then approve.

c) Use Plan Mode before big changes For anything non-trivial, ask Claude to plan first instead of jumping straight to edits:

> Plan how you'd refactor the auth module to use JWT instead of sessions

It’ll research the code and propose a plan you can approve, tweak, or reject — before any files are touched.

d) Write and run tests

> Write unit tests for the utils/formatDate.js file
> Run the test suite and fix any failures

e) Use Git through Claude Code

> Commit these changes with a clear message
> Create a pull request for this branch

f) Pipe things in and out (unix-style)

cat error.log | claude -p "summarize the root cause of these errors"

g) Create custom slash commands for repeated tasks (e.g. a /deploy-checklist command your team reuses).

5. A few essential commands

6. Good early habits

• Start with read-only questions before letting it edit files, to build trust in its judgment on your codebase.
• Keep changes small and review every diff — don’t rubber-stamp.
• If your repo touches sensitive data, ask your admin about permission/hook restrictions before diving in.
• A CLAUDE.md file at your project root (plain markdown describing conventions, gotchas, and dos/don’ts) meaningfully improves output quality — worth setting up early.

Source: Anthropic’s official Claude Code docs — https://docs.claude.com/en/docs/claude-code/overview — for anything more advanced (hooks, MCP servers, subagents, CI/CD integration).