The Claude dashboard is really two separate screens, and only one of them controls your API key

Which Claude dashboard are you actually looking at

Anthropic runs two products that both get called "the Claude dashboard," and most confusion starts with mixing them up.

The chat app lives at claude.ai, plus the desktop and phone apps. This is where you type to Claude, keep chats, and organize work in Projects. It is paid for with a subscription: Free, Pro, Max, Team, or Enterprise.

The Console is the developer side. It started at console.anthropic.com, and Anthropic has been folding it into what it now calls the Claude Developer Platform, so the address and some labels may look different when you log in. This is where API keys, prepaid credits, usage graphs, and spend limits live. You pay for it per use, out of a credit balance you top up.

The two are separate accounts with separate bills, even when you sign in with the same email. A Pro or Max subscription does not put a cent into your Console balance, and Console credit does not unlock anything in the chat app. We covered that split in more detail in why Claude Pro and Max don't come with an API key.

A quick way to tell which one you need:

• If you want to talk to Claude yourself, write emails, or summarize a PDF, you want the chat app.
• If some other software, like a CRM, a booking tool, or a plugin, asked you to "paste your Anthropic API key," you want the Console.
• If you are trying to see why a bill was higher than expected, check which product the charge came from first. Subscription charges and credit top-ups show up as different line items.

The rest of this guide is mostly about the Console. That is the one with real money attached and the fewest plain-English explanations.

The chat app has only three screens worth learning

If you meant claude.ai, there is not much to learn, and that is the point of it. Three areas matter.

The sidebar. Your past chats sit here. Star the ones you return to, because search gets crowded fast. A chat that has run very long gets slower and starts eating into your usage allowance faster. When a conversation has done its job, start a fresh one rather than piling on.

Projects. A Project is a folder with standing instructions and files attached. A body shop owner might make one called "Customer replies" holding the shop's price list, warranty terms, and a note like "write plainly, no exclamation marks." Every chat started inside that Project already knows those things. Projects are the single biggest time saver in the app, and most people never open one.

Settings. Under your name you will find your plan and billing, a usage view showing how close you are to your plan's limits, and the switches for connectors and features. If Claude has "stopped working" partway through a busy afternoon, the usage view usually explains it: you hit the plan limit, and it resets on a timer.

What the chat app does not have is anything to do with API keys. If you go looking for "API" in claude.ai settings and find nothing, you are not missing a menu. It is in the other product.

Setting up the Console in about fifteen minutes

Here is the order that avoids surprises. The step most guides skip is the third one, and it is the one that protects you.

1. Create the account and organization

Sign up at the Console with a business email. You will be asked for an organization name. Use your business name, because invoices are issued to it. If you already have a claude.ai account on the same email, you may be offered the same login, but the billing is still separate.

2. Buy a small amount of credit

The Console runs on prepaid credit. Go to Billing, add a card, and buy the minimum. Anthropic's lowest usage tier has historically needed a $5 purchase to start. For a small business running AI inside one or two tools, $10 to $20 often lasts weeks or months. You can see what yours actually costs after a few weeks on the Usage page, which beats any estimate.

Billing also offers auto-reload, which tops your balance up when it falls below a number you choose. Leave it off for the first month. Once you know your real monthly spend, turn it on with a reload amount close to that number.

3. Make a workspace for each tool and give it a spend limit

Workspaces are folders inside your organization. Each one can hold its own API keys and its own monthly spend limit. Create one per piece of software that will use Claude, named after the software: "CRM," "Website chatbot," "Bookkeeping plugin."

Then open each workspace's limits and set a monthly spend cap. Pick a number that would annoy you but not hurt you, say $15 for a tool you expect to cost $3. If something misbehaves, like a loop or a form spammed by bots, the workspace stops at $15 instead of draining your balance.

4. Create the key inside that workspace

Go to API Keys, choose the workspace, and create a key. Name it after the tool and the date, for example "crm-2026-11." The key, a long string starting with sk-ant-, is shown once. Copy it straight into the tool that asked for it. Do not email it to yourself, put it in a shared doc, or paste it into a chat. If you lose it, delete it and make a new one. That takes ten seconds and is safer than trying to recover it.

5. Make one real request and check it landed

Use the feature in your tool once, such as drafting a reply or summarizing a note, then refresh the Console's Usage page. Within a few minutes you should see a small entry against that key. If you see nothing, the tool is not using the key you think it is.

How to read the Usage and Cost pages without a developer

The Console shows two views that look similar and answer different questions.

Usage counts tokens, the chunks of text Claude reads and writes. Roughly, a token is three-quarters of a word. You will see input tokens (what was sent to Claude: your instructions, the customer's email, any attached notes) and output tokens (what Claude wrote back). Output tokens cost several times more than input tokens, so a feature that writes long answers costs more than one that reads long documents and returns a sentence.

Cost turns those tokens into dollars. This is the page to check monthly. Filter or group it by workspace or key, which is where one-workspace-per-tool pays off. Instead of "you spent $11.40," you see "the CRM spent $2.10 and the website chatbot spent $9.30," and you know where to look.

Three patterns to recognize:

• A flat, low line with small daily bumps on working days is normal. That is a person using a feature when they need it.
• A steady line that runs through weekends and nights means something automatic is calling Claude on a schedule. That may be intended, like a nightly summary, or it may be an automation set to run more often than it should.
• A sudden spike on one day usually means a batch job, a loop, or bots hitting a public form connected to AI. Check which key it came from, then check that tool.

You will also see which model was used. Anthropic sells several, and the largest costs many times more per token than the smallest. If a tool lets you pick, the mid-sized model is usually right for business writing. The current per-token prices are on Anthropic's pricing page. For working out a monthly figure from those prices, our worked example of API cost per month uses the same method, and it carries over to Claude directly.

One screen people overlook is Limits. It shows your rate limits and your organization's monthly spend ceiling for your usage tier. New accounts start in a low tier with a low ceiling. That is a safety net, not a problem, and it rises automatically as you spend more over time.

What most people get wrong on the Console

Most Console problems for small businesses are not technical. They are a handful of habits.

Assuming the subscription covers it. The most common support question in this whole area: "I pay for Claude Max, why is my API key saying no credit?" Because they are different products. You need Console credit, even if it is only $5.

One key for everything. It works, but you lose the ability to see which tool spent what. You also lose the ability to cut one tool off without breaking the others. If you leave a piece of software, you want to delete its key and nothing else.

No spend limit, auto-reload on. That is the combination that turns a bug into a real bill. With a workspace cap in place, the worst case is the cap.

Using the Workbench as a free chat app. The Console has a Workbench for testing prompts. It is genuinely useful for checking how a prompt behaves before you put it in a tool. But every message there is billed at API rates from your credit. For everyday chatting, the subscription app is the better deal.

Keeping keys from tools you stopped using. Any key still active is a key that could be used. Every few months, delete keys you do not recognize or no longer need. The Console shows when each key was last used, which makes the stale ones easy to spot.

Inviting the whole team as admins. You can add staff to the Console with different roles. Most people on your team need no Console access at all. They use the tool that holds the key. Give billing access to whoever pays the bills and keep admin to one or two people.

A five-minute monthly check, and where the key ends up

Once it is set up, the Console needs almost no attention. Put this on the calendar for the first working day of each month:

1. Open Cost, set the range to last month, and group by workspace. Write each number down next to the tool's name. A notes file is fine.
2. Compare against the month before. If anything has doubled with no obvious reason, like a busy season or a new feature you switched on, open that tool and look at what is calling Claude.
3. Check Billing: current balance and whether auto-reload fired more often than you expected.
4. Glance at API Keys and delete anything unused in the last 60 days.
5. If a workspace is regularly hitting its spend cap and the usage is legitimate, raise the cap. If it is hitting the cap for reasons you cannot explain, lower it and find out why.

After two or three months, your notes give you a real monthly AI cost per tool. That number is worth more than any vendor's estimate, including ours.

This is also why holding your own key matters. When software charges you for "AI credits" on top of a subscription, you see one blended number and cannot check it against anything. When the software runs on your key, the Console shows every token it used, and the bill comes straight from Anthropic at Anthropic's prices. Fynl works this way: you paste your Claude or ChatGPT key into the portal, its AI features bill to your own account with no markup, and giving it its own workspace and cap means you can see exactly what it spends. Fynl does not handle SMS or phone calls yet, so if those are what you need AI for, look at a tool built for voice.

Whatever tool you use, check whether it actually needs the API at all before you sign up. Some things people try to do through the Console, like getting Claude to create a Google Doc for you, depend on what the connecting software can do, not on which Claude dashboard you have open.

Questions people also ask

Sources

• Anthropic API documentation: Getting started
• Anthropic API documentation: Rate limits and usage tiers
• Anthropic pricing
• Anthropic Console