How to get an Anthropic Claude API key and connect it to a desktop AI app

To get an Anthropic Claude API key, sign in to the Claude Console at platform.claude.com, open Settings → API keys, click Create key, and copy the key that appears. Anthropic's documentation says the Console shows the full key only once, so store it somewhere safe right away. The API is paid by use and billed apart from a Claude subscription, and Anthropic's help center tells you to buy usage credits on the Console's Billing page before you use it. After that, you connect the key in the desktop AI app you want to run on it.

This guide follows Anthropic's own documentation, help center and Privacy Center as we read them in October 2026, and it links each page so you can check a step against its source. We ran no tests in Anthropic's Console and show no screenshots, and menus can change, so trust the linked page if it differs from this one. The connecting steps use Barside as the worked example. We make Barside, a free desktop app for Windows, Linux and Mac that has no AI of its own and runs on the AI account you connect.

How do you get an Anthropic Claude API key?

Anthropic's quickstart lists two things to have before your first request: a Claude Console account and an API key. Its page Get your Claude API key says where keys are kept: “API keys for the Claude API (also called Anthropic API keys) live in the Claude Console.” It then gives four steps.

1. Sign in to the Claude Console. Go to platform.claude.com and sign in, or create an account if you don't have one yet.
2. Open the API keys page. Go to Settings → API keys.
3. Create a key. Click Create key, name the key, choose an expiration, and set Linked account to yourself or a service account. You can also choose a workspace to scope the key to.
4. Copy and store the key. The Console shows the full key, which starts with sk-ant-, only once, at creation. Copy it and store it somewhere safe. A lost key can't be viewed again in the Console, so you would create a new one.

The same page describes the key types. A personal key acts as you and stops working if you leave the organization, a service account key represents a service account for shared workloads, and a workspace key is a legacy type without an owner. Anthropic's advice is “Use a personal key for your own development, and a service account key for anything shared.” For a desktop app that only you use, that points to a personal key. If the Create key button is disabled, the page says your role may not allow you to create keys there, and an organization admin can change it.

Is the API part of a Claude subscription, and what does it cost?

An API key is billed on its own. Anthropic's help article on paid plans and API billing says “Claude paid plans and the Claude Console are separate products designed for different purposes” and that “the Claude API is billed separately.” The same article adds that “Claude Max and Team plans include monthly credits for the Claude API”, so read it if you hold one of those plans.

Payment works through credits. Anthropic's article How do I pay for my Claude API usage? says “Claude API and playground usage is billed through prepaid usage credits.” Its instruction is “Buy credits before you use the API, and they're applied to your usage according to our current pricing.” Per that page, you log in to the Console with an Admin or Billing role, go to Settings → Billing and click Buy credits. Two more lines from it are worth knowing: “Credits expire one year from the purchase date, and the expiration date can't be extended.” and “Failed requests aren't charged.” The article adds that a request that was on track to succeed is still charged if your app disconnects or times out in the middle of it, and that “All credit purchases are non-refundable.”

What a request costs depends on the model and on how much text goes in and comes out. Anthropic's pricing page lists each model with an input price and an output price per MTok, and its pricing documentation defines the unit: “MTok: Million tokens. $5 / MTok is $5 for every million tokens.” As of October 2026, the pricing page lists Opus 5, in its group of legacy models, at $5 / MTok for input and $25 / MTok for output, with other models at other rates. Barside's Claude connection starts with Claude Opus 5 and lets you choose another model, so check the page for the model you pick.

A worked example with assumed sizes shows the scale. Suppose one exchange sends 2,000 tokens to Opus 5 and gets 500 tokens back. At the prices above, as of October 2026, that is $0.01 for the input and $0.0125 for the output, or $0.0225 in total. The sizes are assumptions for the arithmetic. A real assistant request also carries the list of Barside's tools and any matching note excerpts, and Barside's assistant works in a loop of up to 40 steps per request, each one another exchange with the model. Look at the Usage page in the Console after a few days of use instead of planning around this figure.

How to connect a Claude API key to a desktop AI app

To connect a Claude API key to a desktop AI app, you open the app's AI settings, choose Claude as the AI, and give the app the key you created. Each app words this differently, so follow its own guide. In Barside, the steps are these.

1. Download Barside for Windows, Linux or Mac, and open it.
2. Open the assistant with Ctrl+Shift+A or the Assistant entry in the sidebar. On first use you see Connect an AI to start and a button for each AI company. Settings → AI has the same buttons.
3. Choose Claude. Barside opens Anthropic's key page in your browser.
4. Make a new key there, as in the four steps above, and copy it.
5. Barside finds the copied key on the clipboard, checks it and stores it encrypted.
6. Pick a model with Choose a model, under the buttons. A Claude connection starts on Claude Opus 5, and a model that can see images is the one to pick if the assistant should understand your screen.
7. Ask for something in the box labeled “Ask Barside to do something…”, then press Run.

During the connection, Barside watches the clipboard for up to 10 minutes, looking only for a key from that company. It ignores anything else there and removes the key from the clipboard. Afterwards, the line under the assistant's name shows the AI and the model in use, and the same key runs the AI assistant, AI search over your notes, agents that ask before they act and Quick AI in the search bar.

Two made-up examples show how this plays out. Ines, a freelance translator, creates a personal key named “Barside laptop”, buys a small amount of credits, and tries a first request such as “Turn the checklist in this note into tasks with due dates” (an example prompt). Marco, a bookkeeper who already pays for a Claude plan, skips the key and chooses My plan, which runs Claude Code on his computer, signed in with his plan. He accepts the trade-off Barside's guides describe: web search inside Barside and cloud agents still need an API key.

How to set a spending limit before the first request

Anthropic's rate limits page describes spend limits as a maximum monthly cost for API usage. Each of the Start, Build and Scale usage tiers carries a monthly spend cap, and the page says you can “set your own spend limit below your tier's cap to control costs”. Its steps are short.

1. Go to Settings → Billing in the Claude Console.
2. In the Spend limits section, click Adjust limit, or Set limit if no limit is set yet.
3. Enter a new value. The page notes that “Your spend limit cannot exceed your current tier's cap.”

When usage reaches a limit you set, the page says requests return an error whose message begins “You have reached your specified API usage limits” and states when access resumes. You can raise or remove the limit to restore access sooner. Prepaid credits act as a second ceiling, because the payment article says “If you run out of credits, you can no longer call the API or use the playground until you add more.” That article also describes auto-reload, which “purchases additional credits automatically when your balance falls below a threshold you set”, so with auto-reload on, your credit balance no longer caps what you spend.

Barside adds a limit of its own on your computer. In Settings → AI, the AI spending limit section has the option Limit what Barside's AI spends. Turn it on and pick an amount from US$0.50 to US$10,000 for a day, a week (from Monday) or a month. Barside estimates what each answer costs from the AI company's list prices for your model, warns you at 80%, and stops its AI at the limit until you raise it, turn it off or the count starts again.

Use both limits together. Barside's count is an estimate, each computer counts only its own use, and cloud agents aren't counted at all. The limit in the Console covers the API usage of your whole Anthropic organization, whichever app or computer it comes from.

How to keep the key safe

Anthropic's article API Key Best Practices: Keeping Your Keys Safe and Secure, dated March 16, 2026, explains what is at stake: “Much like a credit card number, if someone obtains and uses your API key, they incur charges on your behalf.” Its advice carries over to a desktop app.

• Share it with no one. “Just as you wouldn't share your personal password, don't share your API key.”
• Give each app its own key. The article recommends separate keys for separate environments. The same idea works for apps: with one key per app, you can delete one without disturbing the rest.
• Set an expiration and rotate. You choose an expiration when you create a key, and the article's advice is to “Regularly rotate your API keys on a consistent schedule (for example, every 90 days)”.
• Watch the usage. “We recommend regularly reviewing logs and usage patterns for your API keys within the Console.”
• Revoke at the first doubt. “If you suspect that your API key may be compromised, we recommend revoking the key immediately.” Then create a new key and connect that one instead.

The article also says Anthropic takes part in GitHub's secret scanning program. When a key turns up in a public repository, “Anthropic automatically deactivates the exposed API key”, which mostly matters to people who keep keys near code.

Does the app's developer see your data when you use your own API key?

With a desktop AI assistant app that runs on your own API key, whether the developer sees your data depends on the route your requests take and on where the key is kept. An app can be built to send each request from your computer straight to Anthropic. It can also be built to pass requests through the developer's servers first, and in that design those servers handle your prompts and your key. The app's own documentation is the place to find out which design it uses.

Anthropic's key safety article treats the key as the thing to protect. It says that when you upload your API key to third-party tools or platforms, such as a web-based IDE, a cloud provider or a CI/CD platform, “you are giving the developer of that tool access to your Claude Console account.” Before you paste a key into any app, look for four answers in its documentation:

• Where the key is stored, and whether it is encrypted.
• Whether requests go straight to Anthropic or through the developer's servers.
• What the app adds to a request, such as note text, files or screenshots.
• Which optional features need your key or your data on a server.

Barside's guides answer them this way.

There is one exception, and it is optional. With cloud agents in Barside Plus, turning on Run scheduled agents in the cloud sends the company, model and key of your AI to Barside's cloud-agents service. That service verifies the key with your AI company and seals it with AES-256-GCM, and its database then holds the sealed bytes and the last 4 characters of the key. The key is used only to run your scheduled agents, and Remove my AI key from Barside deletes it. A cloud run also reads your synced notes on Barside's server, which can read synced notes, so leave cloud agents off if you want every AI request to start from your own computer. The privacy guide explains how agents use your data.

Two limits apply to all of this. When the assistant inspects a window in another app, text from that window can travel inside a tool result, so what is on your screen can end up in a request. Unticking a workspace in Library & briefing → Sources keeps it out of Library answers and agents, but the assistant acts on what you ask, so it can still open a note you point it to. The AI section of the privacy guide has the details.

What do Anthropic's policy pages say about API data?

Once a request reaches Anthropic, Anthropic's policy applies to it. Barside has no say over that policy and cannot pull back data a provider already holds. Two articles in Anthropic's Privacy Center cover the API, and both state the same scope: “This article is about our commercial products such as Claude for Work and the Anthropic API.”

• Retention. How long do you store my organization's data? shows the date July 1, 2026. For API users, it says “we automatically delete inputs and outputs on our backend within 30 days of receipt or generation”, with exceptions it lists: a service with longer retention under your control, a different agreement between you and Anthropic, enforcing its Usage Policy, and complying with the law. It adds that “for Covered Models, we require limited data retention and review as a part of our safety work.”
• Flagged content. The same article says that if a chat is flagged by its automated trust and safety systems as violating its Usage Policy, “We retain inputs and outputs for up to 2 years”, and it keeps “trust and safety classification scores for up to 7 years”.
• Training. Is my data used for model training? shows the date August 18, 2026. It says that by default Anthropic will not use your inputs or outputs from its commercial products to train its models. The exceptions it names are “If you explicitly report feedback or bugs to us”, or if you “otherwise choose to allow us to use your data”.
• Feedback. When you submit feedback, the retention article says “we retain data associated with that submission for 5 years”.

Anthropic's developer documentation adds a stricter arrangement. Its page API and data retention says “Under a ZDR arrangement, Anthropic does not store customer prompts or responses at rest after the API response is returned.” ZDR stands for zero data retention, and the page says you request it for your organization through Anthropic's sales team, so no desktop app can switch it on for you. The page also names Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5 and Claude Mythos 5 as Covered Models that require 30-day data retention, and says ZDR is not available for them unless Anthropic expressly authorizes it. Even then, the page says “Anthropic may retain data where required by law or where it has been flagged by Anthropic's automated trust and safety systems.”

Three cautions belong next to that summary. These pages describe the API and Anthropic's other commercial products, and the training article says Claude Free, Pro and Max, and Claude Code on those plans, are covered in a separate article. That matters if you choose My plan in Barside, which runs Claude Code signed in with your Claude plan. Barside's guides call an API key the safer choice of the two. Policies also change, so read the linked pages and their dates before you rely on any summary, this one included. Finally, neither Anthropic's pages nor Barside's guides say that nothing is ever stored, and this guide makes no such promise.

If you hold other people's confidential material, AI note apps and client confidentiality goes further into what to check before client data goes into any AI feature.

Details and limits

• A connection is required. The assistant sends requests to Anthropic, so it needs the internet. Your voice is still turned into text on your computer.
• Web search is billed by Anthropic. With a Claude key, the assistant searches the web through Anthropic's web search, and Anthropic bills those searches on your account.
• Systems. The same Settings → AI connection works on Windows, Linux and Mac. What the assistant can do inside other apps differs by system, and the AI assistant guide has the table.
• Other AI companies. ChatGPT and Gemini connect the same way, through a key page and the clipboard. OpenRouter is a one-click sign-in with no key to copy.
• What we did not check. The Console steps come from Anthropic's pages as of October 2026, and we did not run them ourselves. Barside's steps come from its own guides.

Three ways to use Claude with Barside

A key tends to cost little when you use AI now and then, and more once long research sessions or daily agents enter the picture. For the numbers behind that choice, see AI research assistant pricing: subscriptions vs paying for the API and AI note-taking app pricing in 2026. If you expect to switch between Claude and other models later, AI notes apps where you bring your own key explains what stays put when you do.

Works well with

Once the key is connected, these guides cover what it can run and how to keep sensitive notes out of it.