Your first LLM API call: from key to response

1. 1

   Create a key and keep it out of your code

   This time your program sends the question instead of a chat window. Use one short, non-personal practice prompt. You can study response handling before making a paid request; check account permissions and API pricing separately from any chat subscription.

   To use an LLM through an API you first need an API key. For the Claude API, sign up for the Claude Console and create a new key there. Every request made with that key is billed to your account, so treat it like a password.

   Don't write the key into your code. Put it in an environment variable called ANTHROPIC_API_KEY instead: the official SDK reads it automatically, so not a single character of the key ends up in your code.

   Where an API key should and shouldn't live

   Where	Verdict	Why
   In code (api_key="…")	No	Leaks the moment you push the repo or share your screen
   Chats, messengers, issue threads	No	Stays in the history, and many people see it
   Environment variable ANTHROPIC_API_KEY	Good for development	The SDK reads it automatically and it never touches your code
   Your cloud's secrets manager	Good for servers and deployment	Access and key rotation are managed in one place

   Check it

   • Do not paste a key into a shell command or chat. The code uses hidden input when the environment variable is absent, keeping that input in the current process.
   • Environment variables still need appropriate process, log and configuration controls. If exposure is suspected, revoke or rotate the key and review usage.

1. 2

   Install the SDK and send a first request

   In a separate folder, run python3 -m venv .venv, activate with source .venv/bin/activate, then install with python -m pip install anthropic (macOS/Linux). Windows uses a different activation path. Record the installed version with python -m pip show anthropic.

   

   If writing it from scratch feels daunting, ask an AI coding assistant, giving it clear conditions as below. Then compare what you get, line by line, with the get-started example in the official docs.

   Review a short anthropic Python SDK example. Never print the key. I will supply a model ID verified in current documentation and my account. Make one attempt with no automatic retries and inspect completion status before displaying text.

   Save as first_call.py and run python first_call.py. Enter a model ID verified in the current documentation and your account. This editorial teaching example is not a recorded model interaction.

   Example codeCopy

   import getpass import os import anthropic def completed_text(message): if message.stop_reason != "end_turn": raise ValueError(f"Response not complete: {message.stop_reason}") text = "\n".join(block.text for block in message.content if block.type == "text") if not text.strip(): raise ValueError("No text in response") return text if __name__ == "__main__": model = input("Model ID from current documentation: ").strip() if not model: raise SystemExit("A model ID is required") key = os.environ.get("ANTHROPIC_API_KEY") or getpass.getpass("API key: ") if not key: raise SystemExit("An API key is required") client = anthropic.Anthropic(api_key=key, max_retries=0, timeout=30.0) try: message = client.messages.create( model=model, max_tokens=1000, messages=[{"role": "user", "content": "List three things a customer-support reply should contain."}], ) print("usage:", message.usage.model_dump()) print(completed_text(message)) except anthropic.APIStatusError as exc: print("HTTP:", exc.status_code, "request_id:", exc.request_id) except anthropic.APIConnectionError: print("Connection failed; check the network before retrying") except ValueError as exc: print(exc)

   Check it

   • Running the script makes a real API request. Verify model access, prices and limits before trying it once.
   • If no answer is shown, inspect the stop reason or error. Check request status and usage before repeating the call.
2. 3

   Read the text, stop reason and usage

   Besides the answer itself, the response carries information your product will need. Make a habit of checking stop_reason and usage from the start.

   

   Check stop_reason before printing content. This single-turn example displays text only for end_turn and holds truncated, refused or other responses. A warning after the answer is too late. Applications with tools need a separate tool_use flow.

   Fields to check first in a response (example values)

   Field	Meaning	Example
   content	A list of blocks holding the answer; the text is in blocks whose type is text	[{type: "text", text: "…"}]
   stop_reason	Why generation stopped	end_turn (finished), max_tokens (cut off at the cap), refusal (declined for safety)
   usage.input_tokens	Tokens in what you sent	31
   usage.output_tokens	Tokens in what came back, including thinking	412

   Test completed_text without any API call. Use fake objects for end_turn, max_tokens and refusal, and confirm that only a completed response yields text.

   Save as test_first_call.py and run python test_first_call.py. These fake responses require neither a key nor a network request.

   Example codeCopy

   from types import SimpleNamespace as NS from first_call import completed_text for reason in ("end_turn", "max_tokens", "refusal"): fake = NS(stop_reason=reason, content=[NS(type="text", text="practice text")]) try: result = completed_text(fake) except ValueError: assert reason != "end_turn" else: assert reason == "end_turn" and result == "practice text" print("3 offline cases passed")

   Check it

   • Check that only the complete response is displayed; truncation and refusal must be held.
   • The table’s 31 and 412 token counts are invented practice values, not measured usage for this prompt.
3. 4

   When it fails, start with the status code

   When a request fails, the API returns an HTTP status code along with an error type and a message. The status code alone usually tells you whether to fix your request or simply try again later.

   This example disables automatic retries with max_retries=0 so one run is easy to reason about. Production retry policies should target transient failures, bound attempts and delays, and consider whether repeating an operation duplicates work.

   Common errors and what to do

   Status	Error type	Meaning	What to do
   400	invalid_request_error	Something is wrong with the request's format or content	Fix the value the error message points to
   401	authentication_error	A problem with the API key (malformed, revoked, expired and so on)	Check the environment variable and the key's status
   429	rate_limit_error	You hit a rate limit or a usage limit	Retry after a pause; if it persists, check your limits in the Console
   500	api_error	An unexpected error inside Anthropic's systems	Retry after a pause; if it persists, contact support with the request_id
   529	overloaded_error	The API is temporarily overloaded by heavy traffic	Retry after a pause

   I ran my first-call code and got this error. What's the cause, and what should I check, in order?
   anthropic.AuthenticationError: Error code: 401 - {'type': 'error', 'error': {'type': 'authentication_error', 'message': 'invalid x-api-key'}}

   A 401 authentication_error means the API didn't accept the key you sent. Check your key setup before the code.

   1) Check which source the code uses. It reads ANTHROPIC_API_KEY first and opens the getpass hidden prompt only when that value is absent. An old key in the environment prevents the new-key prompt from appearing.

   2) Check the key's active status in the Console. Replace revoked keys or keys you suspect have leaked. Never print the value or paste it into an error report or chat.

   3) To use hidden input again on macOS or Linux, run unset ANTHROPIC_API_KEY in the current terminal, then python first_call.py. After the model ID, enter the active key at API key: without surrounding spaces or quotes. The key is not displayed, and you do not put it in the command.

   Check it

   • For a 401, inspect authentication and account configuration instead of repeating the same request. Never attach the full key to a report.
   • A lost connection alone does not prove the server did no work. Check the request ID and usage before retrying.
4. 5

   Work out the cost from usage

   For arithmetic practice only, assume $4 per million input tokens and $20 per million output tokens. The 31/412 and 3,000/600 counts are also invented, not measured usage or a current price quotation.

   The first example costs $0.000124 + $0.00824 = $0.008364. The second costs $0.012 + $0.012 = $0.024. Check units before scaling request volume.

   Use the selected model’s current prices for real work, including caches, tools and other applicable features. Use reported usage, not visible character counts. max_tokens limits output; it is not a complete per-request spending cap.

   Cost practice with fictional prices and token counts

   Request	Input tokens	Output tokens	Cost
   Practice A	31	412	About $0.0084
   A report summary (a long input)	3,000	600	$0.024

   Check it

   • Run the report summary 1,000 times a day and it's $24 a day, or $720 over 30 days. Estimate a month from your expected call volume before you ship.
   • Check real charges on the Console's usage page after a few days of use. Prices can change, so use the current price list every time you calculate.

This is a learning check, not a test you must pass. Use the explanations to understand the ideas.

1. Q1What is wrong with printing the answer before checking max_tokens?

   Write your thoughts (optional)

   Write your thoughts (optional)

   Your writing is not sent anywhere. It disappears when you leave or reload this page, so copy anything you want to keep.

   Read the answer and explanation

   The partial answer has already escaped into use. Inspect the stop reason first and display text only on end_turn in this example.

   I read and understood the explanation

   This is your own review, not an automatic score.

2. Q2At fictional $4/$20 rates, what do 100 calls of 3,000 input and 600 output tokens cost?

   Write your thoughts (optional)

   Write your thoughts (optional)

   Your writing is not sent anywhere. It disappears when you leave or reload this page, so copy anything you want to keep.

   Read the answer and explanation

   Each call is $0.024, so 100 calls cost $2.40. These are invented practice values; actual billing depends on current rates and applicable extras.

   I read and understood the explanation

   This is your own review, not an automatic score.

3. Q3Do three passing offline cases establish successful API integration?

   Write your thoughts (optional)

   Write your thoughts (optional)

   Your writing is not sent anywhere. It disappears when you leave or reload this page, so copy anything you want to keep.

   Read the answer and explanation

   No. They validate local handling of fake responses. Real account access and model behavior require separate evidence from a live request.

   I read and understood the explanation

   This is your own review, not an automatic score.

1. 6

   Try it: record one request

   Run the three offline cases first. Passing them validates local response handling, not real model access, refusal behavior or billing. Record that distinction.

   If you choose a live call, record the model ID, SDK version, time, stop reason, usage and request ID. Exclude keys and customer data. Read the answer separately: transport success does not establish fitness for a customer.

   Explain where the key enters, the condition for displaying text and the units in the cost calculation. Those are the foundations for structured output and code-review work in the next lessons.