How to Get a Claude API Key (and Manage Credits and Billing)

To get a Claude API key, open Claude Console API keys, sign in or create an account, and click Create key. Name it, choose an expiration and linked account, select a workspace, then copy the secret. Anthropic's current key guide documents this flow.

API calls draw from your Console API balance, separately from a Claude subscription. Check that balance before testing.

Already have a key? Test it, buy API credits, or diagnose an error.

How to get your Claude API key

The steps below follow Anthropic's creation instructions, checked September 21, 2026. “Claude API key” and “Anthropic API key” refer to the same direct-provider credential.

1. 1
   Sign in to the Claude Console

   Go to the Claude Console and sign in or create an account. API availability depends on your location, so check Anthropic's supported countries and regions if you cannot register.

2. 2
   Create a named key and choose its owner

   Open API keys settings, click Create key, and give it a descriptive name, such as local-development.

   Set Linked account to yourself for a personal development key. For a shared app, CI pipeline, or production service, use a service account. Ask an organization admin to create that identity if needed. Identity-backed keys stop working when their linked identity loses the applicable access; legacy workspace keys can outlive their creator's membership.

   If Create key is disabled, ask your admin to check your role. Buying credits will not fix a permission restriction.

3. 3
   Choose a workspace and expiration

   For your first integration, scope the key to a single workspace. This avoids having to send a workspace-selection header with each request.

   Choose a lifetime appropriate for the task. Anthropic supports preset and custom expirations, subject to your organization's policy. Expiration is fixed at creation, so changing it later requires a replacement key.

4. 4
   Copy and store the secret

   Copy the full key into a secrets manager before closing the creation dialog. The Console shows the secret only once; it starts with sk-ant-. If you lose it, create a replacement key.

Buy Claude API credits and configure billing

For most self-serve organizations, Claude API billing uses prepaid credits. Those credits cover API calls, playground use, and Console-billed Claude Code. Creating a key and buying credits are separate tasks.

Buy credits in the correct organization

Sign in with an Admin or Billing role, then:

1. Open Console Billing.
2. Select Buy credits.
3. Enter the amount and confirm the purchase.
4. Check the available balance on the same page.

Purchased credits are available immediately. Fund the organization associated with the key. If you belong to several organizations, check which one is selected before purchasing. When its credits run out, API and playground use stop until more are added.

Auto-reload keeps a balance funded, not spending capped

On the Billing page, choose Edit under Auto-reload, enable it, and configure the minimum balance and reload setting shown in the Console. Auto-reload buys credits when the balance falls below your threshold.

That keeps the app funded. To control how much it spends, set a separate monthly spend limit before enabling unattended usage.

Receipts, invoices, and expiration

For receipts, open Settings, then Billing, then Invoice history, and choose Download or View. Each credit purchase generates a receipt, including auto-reloads. Expired grants can appear in that history too, without being new charges.

Purchased API credits expire one year after purchase and are non-refundable. Buy against expected usage. Promotional credits can have their own expiration.

Some organizations arrange monthly postpaid invoicing through Sales. Anthropic's payment guide describes those organizations as receiving a Stripe invoice at calendar-month end.

A Claude subscription is not an API credit balance

Claude Pro and Max subscription access is separate from direct API access. Pro does not include Console API usage. An environment API key causes Claude Code to use API pay-as-you-go billing, even when you are logged in with a Claude subscription.

There are three billing paths you may encounter:

Paid-plan usage credits are charged separately from the subscription. They follow their own expiration rules, rather than the Console's one-year rule. With Requesty's BYOK path, your own provider account retains its billing relationship.

Check the credential your tool is sending, then inspect the account that credential bills. A positive balance elsewhere does not fund this API request.

Test your Claude API key

These examples use a workspace-scoped Console key and the active claude-haiku-4-5 model. The prompt asks for a short hello with a 64-token output ceiling. A few dozen tokens at Haiku 4.5's rates of $1 per million input tokens and $5 per million output tokens cost well under a cent.

Set the key in your current terminal session on macOS or Linux:

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

On Windows PowerShell, use the documented environment-variable form:

$env:ANTHROPIC_API_KEY="your-api-key-here"

An .env file alone does not load values into the process. Configure a loader if you choose that approach, and exclude the file from Git.

If you have configured a gateway before, run the cleanup commands in the note below before either SDK test.

curl --include https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-4-5",
    "max_tokens": 64,
    "messages": [
      {"role": "user", "content": "Reply with a short hello."}
    ]
  }'

pip install anthropic

from anthropic import Anthropic
 
client = Anthropic()
message = client.messages.create(
    model="claude-haiku-4-5",
    max_tokens=64,
    messages=[{"role": "user", "content": "Reply with a short hello."}],
)
 
for block in message.content:
    if block.type == "text":
        print(block.text)
print(message.usage)

python test_claude.py

npm install @anthropic-ai/sdk

import Anthropic from "@anthropic-ai/sdk";
 
const client = new Anthropic();
const message = await client.messages.create({
  model: "claude-haiku-4-5",
  max_tokens: 64,
  messages: [{ role: "user", content: "Reply with a short hello." }],
});
 
for (const block of message.content) {
  if (block.type === "text") {
    console.log(block.text);
  }
}
console.log(message.usage);

node test_claude.mjs

export ANTHROPIC_WORKSPACE_ID="your-workspace-id-here"

$env:ANTHROPIC_WORKSPACE_ID="your-workspace-id-here"

-H "anthropic-workspace-id: $ANTHROPIC_WORKSPACE_ID"

import os
 
client = Anthropic(
    default_headers={
        "anthropic-workspace-id": os.environ["ANTHROPIC_WORKSPACE_ID"]
    }
)

Set spending limits and check your usage tier

A request can fail even with credits available. Your balance, chosen budget, Anthropic's tier cap, and request throughput are separate controls.

Set your own monthly budget

Open Console Billing. Under Spend limits, click Adjust limit, or Set limit if none exists. Your chosen limit cannot exceed your tier's cap.

Choose the amount you are prepared to spend. Adding prepaid credits does not remove this limit, and enabling auto-reload does not replace it.

Current monthly tier caps

Anthropic consolidated its standard usage tiers into Start, Build, and Scale on June 26, 2026. The following monthly spend caps were checked September 21, 2026:

New organizations may begin in Evaluation, with lower limits while account history is established.

Placement and upgrades depend on usage history and account standing. Buying a particular credit amount does not guarantee a higher tier. To request a higher cap or throughput, use Request rate limit increase on the Rate limits page.

Throughput limits measure requests per minute, input tokens per minute, and output tokens per minute. Short bursts can exhaust capacity even when a minute-wide average looks low. Check your account's effective limits in the Console.

Separate environments with workspaces

Workspaces separate apps, environments, or teams while keeping centralized billing. Use separate keys and workspaces for development and production when you need that separation.

For budget isolation, use a non-default workspace: workspace spend and rate limits can be lower than the organization's, but cannot be set on the Default Workspace. Organization-wide limits still apply. Workspaces share the organization's purchased-credit balance.

Check usage and understand the bill

Open Usage for activity and Cost for charges. Usage data can be filtered or grouped by API key, workspace, and model. Reports typically receive data within five minutes, sometimes longer.

Assume 2,000 uncached input tokens and 1,000 output tokens, with no tools, caching, or other pricing modifiers. At Haiku 4.5's standard rates of $1 input and $5 output per million tokens:

The Messages API is stateless: later turns must include earlier conversation messages again. That context becomes billable input again, with eligible caching changing the rate. Longer chats therefore increase the input bill even when each new user message is short.

Fix common Claude API key errors

First isolate the direct API call using the curl test above. If it succeeds but an integration fails, investigate that integration's credential and request configuration before buying more credits or replacing a working key.

The table combines Anthropic's API error definitions, authentication rules, spend and rate-limit behavior, and low-credit diagnosis.

Anthropic's issue tracker records a low-credit 400 response, while its current API reference defines 402 billing errors. Read the message to distinguish a depleted balance from another payment problem.

For a tier-cap 429, error.details.error_code is enforced_spend_limit_reached, and the response sends no retry-after header. Access remains paused until the cap is raised or 00:00 UTC on the first day of the next month. Even automatic SDK retries fail while that cap remains reached.

For support, save the error body and request-id, never the secret. The response organization and workspace headers also help verify which account handled the request.

Rotate keys and finish the setup

Follow Anthropic's key-security guidance: keep secrets out of Git, use secure secret storage, and separate development, testing, and production credentials.

For routine rotation, create a replacement, update your secrets and deployments, verify a successful request, then disable or delete the old credential. Disable is reversible; Delete is permanent, and expired keys cannot be reactivated. If you suspect a leak, disable or delete the compromised key immediately.

Use one Requesty key for Claude and other models

Use an Anthropic key for direct Claude API access. When your app needs multiple providers, we offer one Requesty key for Claude and 600+ models in total, with OpenAI-compatible and Anthropic-compatible endpoints.

Set a monthly spend cap for each Requesty API key and track costs by key, model, provider, and user. For multiple apps or environments, label your Requesty keys so you can attribute usage to the right workload.

For Requesty-managed access, you need only a Requesty key. Set REQUESTY_API_KEY, then use this Anthropic-compatible Messages example. Browse Claude and other providers in our model catalog.

macOS/Linux:

export REQUESTY_API_KEY="your-requesty-key-here"

Windows PowerShell:

$env:REQUESTY_API_KEY="your-requesty-key-here"

Using the Python SDK installed above, save this as test_requesty.py:

import os
from anthropic import Anthropic
 
client = Anthropic(
    api_key=os.environ["REQUESTY_API_KEY"],
    base_url="https://router.requesty.ai",
)
message = client.messages.create(
    model="anthropic/claude-haiku-4-5",
    max_tokens=64,
    messages=[{"role": "user", "content": "Reply with a short hello."}],
)
 
for block in message.content:
    if block.type == "text":
        print(block.text)
print(message.usage)

python test_requesty.py

The Anthropic SDK base URL is https://router.requesty.ai; the SDK appends the Messages path.

Add your Anthropic key once with Bring Your Own Keys to keep using your Console billing. Configure a policy that uses your Anthropic key, set the request's model to policy/your-policy-name, and authenticate with your Requesty key for one endpoint and per-key caps.

For Requesty-managed access, our pay-as-you-go pricing adds 5% to provider token costs, with no seat fees or minimum spend.

One credential, explicit budgets

Start with one Requesty key, set its monthly cap, and use Claude alongside other providers through the same gateway. Create your Requesty account: it is free to get started, with no credit card required.

Sources

Product details and prices checked September 21, 2026.

• Anthropic: current API key creation and authentication
• Anthropic: API payments, auto-reload, and credit expiration
• Anthropic: current tiers, spend caps, and rate limits
• Anthropic: Haiku 4.5 model and pricing
• Versioned Python client and published JavaScript client
• Requesty: spend limits, BYOK, and pricing