A Claude dashboard tutorial that starts with the spending cap, not the API key

Which Claude dashboard this tutorial covers

This walks through the developer Console, the screen where you buy API credit, create API keys, and set limits. It is not the chat app at claude.ai. Those are two separate screens with separate bills, and a Claude Pro or Max subscription does not pay for API usage. If that split is new to you, read why the Claude dashboard is really two separate screens first, then come back.

You need the Console when a tool asks you to "bring your own Claude API key." That might be a CRM, a bookkeeping add-on, or a website chatbot. The tool sends your requests to Anthropic, and Anthropic bills you directly at its published rates. The Console is where that bill lives.

Anthropic has renamed and moved parts of the Console over time, so a menu label may not match a screenshot from last year. The pages a small business actually uses have stayed the same in substance:

• Billing: prepaid credit and auto-reload.
• Limits: your usage tier, rate limits, and spending ceilings.
• Workspaces: separate containers for keys, each with its own limits.
• API keys: the strings you paste into tools.
• Usage and Cost: what was spent, by key, model, and day.
• Workbench: a test bench where you can run a prompt yourself.
• Members: who else can log in and what they can touch.

The rest of this guide covers those pages in the order you should use them, which is not the order they appear in the sidebar.

The setup order that keeps a bad tool from costing you real money

The usual tutorial opens with "click API keys, click Create Key." That works. But the moment the key exists and is pasted somewhere, it can spend whatever your account allows. If the tool has a bug that loops, or someone copies the key out of a shared document, the only thing standing between you and a big bill is whatever limit you set earlier. If you set none, it is the maximum your tier permits.

So reverse it. Build the fence, then open the gate:

1. Add a small amount of credit. Enough for a few weeks of what you expect. You can always add more.
2. Decide on auto-reload. Leave it off for the first month.
3. Create a workspace for the tool and give it a monthly spend limit.
4. Create the API key inside that workspace, named after the tool.
5. Test once in the Workbench, then paste the key into the tool and check the Usage page the next day.

Here is what that looks like with numbers. Priya runs a two-person photography studio. She connects her client portal's AI features to her own Claude key so it can draft replies to inquiry emails. She buys $20 of credit, leaves auto-reload off, and creates a workspace called portal-inbox with a $15 monthly limit. The worst case in her first month is now $15, and the credit balance itself caps it at $20 even if she gets the limit wrong. If she had started with "Create Key" and turned on auto-reload, her worst case would have been her tier's ceiling.

None of those amounts are a recommendation. They are an example of the shape: credit slightly above the cap, cap slightly above what you expect to spend.

Billing is prepaid credit, and auto-reload is the setting that matters

The most common misunderstanding about the Console is that it works like a subscription. It doesn't. You buy credit up front, and each request draws it down at the per-token price of whichever model the tool uses. When the balance hits zero, requests fail until you top up. Current per-model prices are on Anthropic's pricing page.

That prepaid design is a safety feature, but only if you leave auto-reload off. Auto-reload buys more credit automatically when your balance falls below a threshold you pick. It's convenient once you trust your usage. Turned on in week one, it removes the one hard stop the prepaid model gives you.

What to do on the Billing page:

• Add a modest first purchase. Note that your usage tier, covered below, is partly based on how much credit you have bought over time. A tiny first purchase can mean a lower tier and lower limits. For a single small business tool, that rarely matters.
• Leave auto-reload off for at least one full billing month. Look at what you actually spent, then decide.
• If you do turn it on later, set the reload amount small. A $10 top-up that fires twice in a week is a signal you can act on. A $200 top-up that fires once is a signal you will miss.
• Check where receipts go. If your bookkeeper needs them, make sure the billing email on the account is one that person can see.

What is not true: running out of credit does not "pause" your tool gracefully. Most tools just show an error or quietly stop drafting. If a tool's AI features suddenly stop working, check your credit balance before you email that tool's support.

Workspaces and spend limits do the real protecting

Your account has an organization-level ceiling set by your usage tier. Anthropic places every account in a tier, and each tier carries a maximum monthly spend and rate limits. You move up by buying more credit over time and waiting. The current thresholds change, so read them on Anthropic's rate limits page instead of trusting a number from a blog post, including this one. Your own tier and ceiling appear on the Limits page in the Console.

A tier ceiling is a backstop, not a budget. It is usually far above what a small business tool should spend. The setting that matches your actual budget is the workspace spend limit.

A workspace is a container inside your account. Keys live in a workspace, and you can give each workspace its own monthly spend limit and its own rate limits. When the workspace hits its limit, keys in it stop working until the month resets or you raise the limit. Other workspaces keep running.

How to set it up:

• Create one workspace per tool or per job: crm-email, website-chat, bookkeeping. Use names you'll still understand in six months.
• Set a monthly spend limit on each. Start low. Raising it takes a minute. Explaining a bill takes longer.
• Leave the Default workspace empty, or use it only for your own testing.

Splitting by workspace also makes the Cost page readable later, because spending is broken out per workspace and per key. The case for giving every tool its own key is covered in how the usage dashboard shows what each API key spent, so this guide won't repeat it. The short version: one key per tool, one workspace per budget.

Creating the key and handing it to a tool

Now create the key, inside the workspace you just made. The Console shows the full key once, at creation. It starts with sk-ant-. Copy it straight into the tool that needs it. If you lose it, don't hunt for it. Create a new one and delete the old one.

Habits that save trouble:

• Name the key after the tool and the date, like portal-inbox-2026-11. When you rotate it next year, you will know which one to delete.
• Paste it only into the tool's own settings screen. Not into a contact form, a shared Google Doc, a group chat, or an email to "the guy who's setting it up." If a contractor needs it, invite them as a member of the Console with limited permissions, or paste it in yourself.
• Never put it in your website's code. Anything in a public web page can be read by anyone who visits. A chatbot widget that needs a key should store it on the vendor's server, not in your page.
• Disable before you delete if you're not sure a key is still in use. If something breaks, you'll know which tool used it and can turn it back on.

For context, this is the arrangement Fynl uses: its AI features run on your own Claude or ChatGPT key, you paste the key into the portal's settings once, and Anthropic bills you directly with nothing added on top. The same rules apply to Fynl as to any other tool. Give it its own workspace and key, and cap that workspace. If the tool you're connecting resells AI usage instead, you won't need a Console key at all. You'll be paying that tool's markup instead, which is a different tradeoff.

Once the key is in the tool, use the tool normally for a day. Then open the Usage page and confirm you see activity under that key. Seeing nothing usually means the key was pasted wrong or the tool is set to a different provider.

Use the Workbench to test before you blame the tool

The Workbench is the most underused page in the Console, and for a non-developer it is the most useful. You type a prompt, pick a model, optionally add a system prompt (standing instructions such as "you are replying on behalf of a body shop; be brief"), and see the response. Workbench runs use your API credit, so they show up on your bill like any other request. Each test costs very little.

Three things to use it for:

• Separating a tool problem from a model problem. If a tool's AI drafts read badly, paste the same customer email and similar instructions into the Workbench. If the Workbench answer is good, the tool's instructions or settings are the problem. If it's equally bad, your instructions need work, whichever tool you use.
• Comparing models on your own work. Run the same real task, such as summarizing a long quote request, on a cheaper and a more capable model. Many tools let you choose which model they call. If the cheaper one is good enough for your inbox, that choice can cut your spend substantially. Check the per-token prices on the pricing page to see by how much.
• Getting a feel for length. Long pasted documents and long answers cost more because pricing is per token. Seeing how much a typical request uses helps you set a realistic workspace limit.

You can ignore the "Get code" button. It produces code for developers who are building their own app. If you are pasting a key into an existing tool, you don't need it.

A five-minute monthly check, and what the dashboard can't tell you

Once things are running, the Console needs very little attention. Once a month, check:

1. Cost page: total spend, and spend per workspace. Anything that jumped compared with last month?
2. Limits: did any workspace hit its cap? If so, either raise it on purpose or find out why.
3. Billing: credit balance, and whether auto-reload fired more often than you expected.
4. API keys: anything you no longer use? Delete it.
5. Members: anyone who left, or a contractor whose job is done? Remove them.

What the Console won't show you is why something was spent. It knows that the crm-email key used a certain number of tokens on Tuesday. It does not know that it was 40 drafted replies to wedding inquiries, or that one runaway automation retried the same email all afternoon. That information lives in the tool that made the requests. If a workspace's cost jumps, the Console tells you which tool to look at. That tool's own activity log tells you what happened.

So when you choose tools that take your own key, prefer ones that show you what their AI did: a history of drafts, summaries, or automation runs. Without that, you can see that money was spent and which key spent it, but not what it paid for.

One more limit worth knowing. A Claude key only covers text work: drafting, summarizing, sorting, answering. It won't make tools send texts or answer the phone. Whether a given tool can do those depends on the tool. Fynl, for example, does not do SMS or phone calls yet.

Questions people also ask

Sources

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