How to Get a Claude (Anthropic) API Key
Jennifer Tee10 min read
On this page
- What you need before you start
- Is a Claude API key free?
- Does Claude Pro give you an API key?
- How do you create an Anthropic API key?
- What does the Claude API cost?
- Which Claude model should you pick for support?
- Connecting the key to your support tool
- What the spend cap actually stops
- Before you walk away
- Common questions
Getting a Claude API key takes about a minute. Sign in at platform.claude.com, open Settings then API keys, click Create key and name it, then copy the key somewhere safe before you close the dialog. The key starts with sk-ant- and the Console shows it in full exactly once, so a key you fail to copy is a key you have to recreate. That's the whole procedure, and Anthropic's own documentation is the place to check it if the Console ever moves a button.
What you do with the key afterwards is the more interesting question. If you're setting this up so a support tool can run on your own account rather than a vendor's, the case for holding your own provider account explains why that changes your bill. This piece covers the mechanics and the two decisions that actually affect what you spend.
What you need before you start
You need an Anthropic account and, if your company already has one, permission inside it. Everything else happens in the browser.
Anthropic accounts are organized into workspaces, and keys are scoped to them. If you're joining an existing company account rather than creating your own, the Create key button can be grayed out, which means you don't have permission in that workspace. Anthropic's documentation is direct about the fix: ask an organization admin to grant you access, or to create the key for you.
Billing is separate from key creation. The Claude Console is also where you set up billing, and it's worth doing that in the same sitting. A key with no billing behind it stops being useful the moment something calls it.
Is a Claude API key free?
The key costs nothing to create, and using it does. Anthropic's pricing page says new users get "a small amount of free credits to test the API", which is a trial balance rather than a standing allowance. Once it runs out you're on paid usage, and there's no recurring free tier underneath it.
That's worth knowing if you're weighing providers rather than committing to one. Google gives Gemini an ongoing free tier, though it comes with a data term that rules it out for customer conversations, which is why the Gemini walkthrough spends most of its time on that trade. Anthropic's arrangement is simpler to reason about: trial credits to evaluate, then you pay for what you use.
It also explains the sites offering free or heavily discounted Claude keys, some of which rank well for this search. A key you didn't create sits inside somebody else's organization, which means their billing, their spend cap, and their visibility over every conversation you send. For a support agent handling real customer messages that's the wrong trade at any price. Create your own and put a small amount of credit behind it.
Does Claude Pro give you an API key?
No, a Claude Pro or Max subscription doesn't include API access. Those plans pay for the chat app at claude.ai, while API usage is bought separately through the Claude Console. The same login gets you into both, which is where the confusion starts, but your subscription doesn't carry any API credit with it and Anthropic's API docs never list a consumer plan as a route to a key.
What you do get as a new API account is a trial balance. Anthropic's pricing FAQ says new users receive "a small amount of free credits to test the API", and after that it's usage billing set up on your own Console billing page. So the sequence is sign in, create the key, then add a payment method, and a key that exists is not the same thing as a key that can bill.
Worth checking your own billing screen rather than trusting a number in a tutorial, since account requirements change and this one changes often.
How do you create an Anthropic API key?
You create one in the Claude Console, and it takes four steps with something to verify at each one.
-
Sign in to the Claude Console. Go to platform.claude.com and sign in, or create an account if you don't have one. You'll know it worked when the Console dashboard loads rather than the marketing site.
-
Open the API keys page. Go to Settings then API keys. The page lists any keys the workspace already has, so an empty list here is normal on a new account.
-
Create a key. Click Create key, give it a name, and optionally scope it to a workspace and set an expiration, then confirm with the Add button. A name like "support agent, production" beats "key 1" the first time you have three of them.
-
Copy and store the key. The Console shows the full
sk-ant-value once, at creation. Paste it into your secrets manager before you dismiss the dialog, then confirm it's saved by reading it back from wherever you stored it.
Name and expiration are both worth using. A key scoped to one purpose can be revoked without breaking everything else you've connected, and an expiry date turns a forgotten credential into one that retires itself.
What does the Claude API cost?
You pay per million tokens, with no subscription and no minimum spend. Haiku 4.5 costs $1 for a million input tokens and $5 for a million output, Sonnet 5 costs $2 and $10, Opus 5 costs $5 and $25, and Fable 5 costs $10 and $50, all checked against Anthropic's pricing documentation on 27 August 2026.
Rates only mean something once you turn them into a conversation. Anthropic's own worked example puts a support conversation at roughly 3,700 tokens, which is about a third of a cent on Haiku 4.5, or near $37 for ten thousand of them. Bundled support tools charge $0.75 to $1.50 for each conversation their AI resolves, and that gap is the whole argument for holding your own key.
Which Claude model should you pick for support?
For most support traffic, start at the cheap end and move up only where answers get worse. The spread between Claude models is large enough that this decision affects your bill more than anything else in this article.
Haiku 4.5 is the obvious starting point for support, because the work is mostly reading your help content and writing a short reply. Sonnet 5 is the sensible step up when answers need more reasoning. Opus 5 and Fable 5 are built for long agentic work, not for answering a question about a refund policy. Paying frontier rates for "what are your opening hours" is money you won't get back.
Output is where the spending happens, since every model charges four to five times more for what it writes than for what it reads. A support agent that answers in three sentences costs meaningfully less to run than one that produces an essay per ticket.
Rates only turn into a bill once you multiply them by your own traffic, so it is worth estimating the monthly cost at your ticket volume before you settle on a model.
Support traffic also gets a discount most workloads don't, because the front of every request is identical. Your agent instructions, your escalation rules and the help content you attach don't change from ticket to ticket, and Anthropic bills that repeated portion at a tenth of the normal input price once it's cached. The rate limits documentation puts the cache read at "10% of base input token price". Anything that can wait rather than answer live goes through the Batch API instead, which takes 50% off input and output together.
Cached tokens are free of your rate limit
Anthropic states that "for most Claude models, only uncached input tokens count toward your ITPM rate limits". Its own worked example: a 2,000,000 token per minute limit with an 80% cache hit rate processes 10,000,000 tokens a minute. Caching cuts the bill and raises the ceiling at the same time.
Connecting the key to your support tool
Paste the key into whatever is going to use it, then send one real question through and watch for a reply. That round trip is the only proof that the key, the billing, and the model choice all work together.
If you're wiring it up yourself, Anthropic's SDKs read an environment variable called ANTHROPIC_API_KEY automatically, and direct HTTP requests pass the key in an x-api-key header. If you're connecting it to a support platform instead, the key usually goes into a settings field and the platform handles the rest.
Optlo is one of the tools that works this way. You connect your own Anthropic, OpenAI, or Gemini account, then use a fast model for routing and a more capable one for replies instead of paying one rate for everything. If the routing model is going to be an OpenAI one, that's a separate console with its own prepaid billing to set up.
What the spend cap actually stops
Your organization already has a monthly ceiling whether you set one or not, and it belongs to your usage tier rather than to you. Anthropic places organizations on a tier automatically based on usage history, and each tier carries a cap:
| Usage tier | Monthly spend cap |
|---|---|
| Start | $500 |
| Build | $1,000 |
| Scale | $200,000 |
| Custom | no cap, arranged with an account team |
The behavior at the ceiling is the part to plan around. Anthropic's documentation says that once you reach the cap, "API usage pauses until the next month unless you request a higher limit". On a support agent that reads as replies stopping partway through a month, so a new organization on the Start tier should know its number before connecting anything customer-facing.
Rate limits move with the tier too, and they're generous enough that spend is the constraint you'll meet first.
You can set your own limit below the tier cap on the Billing page, and that's the one worth doing today. If your support agent shares an organization with other projects, put it in its own workspace and cap that instead: Anthropic lets you set custom spend and rate limits per workspace, so a runaway experiment somewhere else cannot spend the budget your customer replies depend on. The one exception is the default workspace, which cannot carry its own limit.
Before you walk away
A leaked key is the risk people worry about, but a workflow that retries too eagerly is the one that actually shows up on invoices, and a cap costs nothing to configure.
Anthropic publishes a worked example for exactly this shape of workload, and it's the number worth carrying away. Their pricing documentation puts 10,000 support tickets at around 3,700 tokens per conversation on Haiku 4.5, for a total near $37. That's the whole argument for holding your own key in one figure: the model cost on a real support volume is small, and what you save by paying it directly is the markup somebody else would put on top.
Then work out what the traffic will really cost you. A key is only half the picture, and setting tokens at cost against bundled per-resolution pricing is the comparison most teams are really making when they come looking for a key.
Common questions
How do I get a Claude API key?
Sign in to the Claude Console at platform.claude.com, go to Settings then API keys, click Create key, give it a name, and copy the key before closing the dialog. The key begins with sk-ant- and Anthropic shows it in full only once. You can optionally scope the key to a workspace and give it an expiration date at the same time.
Is a Claude API key the same as an Anthropic API key?
Yes, they're two names for the same credential. Anthropic's documentation calls them "API keys for the Claude API (also called Anthropic API keys)", and both phrases refer to the sk-ant- key you create in the Claude Console. Any guide referring to console.anthropic.com is describing the same place under its former address.
Can I recover a Claude API key I forgot to copy?
No, and there is no workaround. The Console displays the full key once at creation and there is no way to view it again. Anthropic's Admin API can list your organization's keys and return a partially redacted hint, but it never returns the secret value and cannot recover a lost key. Create a replacement key and update whatever was using the old one.
Why is the Create key button grayed out?
You most likely don't have permission to create keys in that workspace. Anthropic's documentation says to ask an organization admin either to grant you access or to create the key for you. This catches people joining an existing company account rather than setting up their own.
Which Claude model should I use for customer support?
Start with Haiku 4.5, the cheapest tier at $1 per million input tokens and $5 per million output. Support work is mostly retrieving your help content and writing a short answer, which doesn't need a frontier model. Move to Sonnet 5 for questions that need more reasoning, and reserve Opus 5 or Fable 5 for genuinely hard work, since paying frontier rates for routine questions adds cost without improving the answer.
Does a Claude Pro subscription include API access?
No, a Claude Pro or Max subscription doesn't include API access. Those plans pay for the chat app at claude.ai, while API usage is billed separately through the Claude Console. Your login works in both places, which is why people expect one to cover the other, but a consumer subscription carries no API credit. New API accounts get a small trial balance, and after that you add a payment method in the Console and pay for what you use.
Do I need a paid plan to use the Claude API?
The Claude API bills separately from consumer Claude subscriptions, and billing is set up in the Claude Console alongside your keys. Creating a key and setting up billing are two different steps, so it's worth confirming both before you point anything at the key. Check your own billing screen rather than trusting a figure quoted in a tutorial, since account requirements change.
