Install the Anthropic SDK

Two install commands, one env var, and a one-liner that gets Claude to say hi back. That's the whole job today.

This is post 9 of 10 in the Setup Toolbox series. The Anthropic SDK is the official client for the Claude API, and it ships in two flavors: Python (anthropic) and Node (@anthropic-ai/sdk). Same API surface in both. We'll install both and verify each.

You'll need an API key from console.anthropic.com. The free credits cover this verify step many times over, so don't worry about burning through them.

Python

Install it via uv inside a project:

# add the anthropic sdk to your project

uv add anthropic

Want a quick one-off with no project around it? Use uv tool install or uv run:

# run a script with the sdk available, no project setup

uv run --with anthropic python -c "from anthropic import Anthropic; print(Anthropic().messages.create(model='claude-haiku-4-5-20251001', max_tokens=20, messages=[{'role':'user','content':'hi'}]).content[0].text)"

Still on an older pip project? pip install anthropic works fine.

Node

Install it via npm inside a project:

# add the anthropic sdk to your project

npm install @anthropic-ai/sdk

pnpm and yarn behave identically (pnpm add @anthropic-ai/sdk, yarn add @anthropic-ai/sdk).

Configure the API key

The SDK reads ANTHROPIC_API_KEY from the environment. Don't hardcode it. For development, drop it in a .env file and add .env to your .gitignore:

# create .env with your api key

echo 'ANTHROPIC_API_KEY=sk-ant-...' >> .env

In Python, load it with python-dotenv, or just export ANTHROPIC_API_KEY=... in your shell. In Node, load it with dotenv or lean on the shell env. For production, use whatever secret manager your platform gives you (AWS Secrets Manager, GCP Secret Manager, Vercel env vars). Never commit the key.

Verify (Python)

# verify python sdk with a one-line call

from anthropic import Anthropic; print(Anthropic().messages.create(model="claude-haiku-4-5-20251001", max_tokens=64, messages=[{"role":"user","content":"hi"}]).content[0].text)

Save it to verify.py, then run uv run python verify.py. You should see Claude's reply printed.

Verify (Node)

// verify node sdk with a one-line call

import Anthropic from "@anthropic-ai/sdk"; const r = await new Anthropic().messages.create({ model: "claude-haiku-4-5-20251001", max_tokens: 64, messages: [{ role: "user", content: "hi" }] }); console.log(r.content[0].text);

Save it to verify.mjs, then run node verify.mjs.

Common gotchas

• Wrong env var name: it's ANTHROPIC_API_KEY, not CLAUDE_API_KEY or ANTHROPIC_KEY. The SDK only reads the official name.
• Model IDs change: pin the full ID with the date suffix (claude-haiku-4-5-20251001) when you want stability. The shorter alias (claude-haiku-4-5) follows the latest snapshot and can shift behavior under you.
• Rate limits hit fast on free credits: a tight loop of test calls will trip the per-minute limit. Add time.sleep(1) or read the usage field to back off.
• Streaming differs from one-shot: .messages.create is one-shot. For token-by-token streaming, use .messages.stream (Python) or .messages.stream (Node). Different return type, you iterate over deltas.
• Prompt caching: turn it on explicitly with cache_control blocks. The SDK won't cache for you; you tag what's cacheable. Worth it for any prompt over 1k input tokens that you reuse.

Once verify.py or verify.mjs prints a reply, you're ready to build against Claude. Most production code differs from that verify script in only two places: the message list and the error handling. Post 10 closes out the toolbox.