How to Build Useful Mini Apps with Claude Artifacts
How Claude artifacts work today, how to describe a tool so the first version is useful, and five practical builds with prompts you can paste
A lot of the small tools people pay a monthly subscription for are a form, a list and a bit of math. A quote calculator, a tracker, a checklist that remembers what you ticked. Claude can build those for you as artifacts, right next to the conversation, from a plain description. I use them constantly for the small tools that are not worth a real app.
Here is how artifacts work today, how to describe one so the first version is useful, and five builds worth trying.
What an artifact is now
An artifact is something Claude makes that you would put in front of someone: a document, a design, a deck, a dashboard or a small interactive tool. It opens beside the chat, you edit it by talking to Claude, and it is saved in the Artifacts tab in the sidebar so you can find it from any conversation.
What you can do depends on your plan. According to Anthropic's help article, anyone can create artifacts in a chat. Saving data between visits, connecting your apps, and starting from templates are on paid plans. One setting trips people up: artifacts need Cloud code execution and file creation turned on in Settings, under Capabilities.
If you made artifacts before September 16, 2026, those are now called legacy artifacts. They still work and can still be published, but new ones follow the current model described here.
Describe the job the tool does
The difference between a toy and a tool is almost always in the prompt. A good artifact prompt answers four questions:
- Who uses it and when. "I use this on calls with clients" leads to a different layout than "I check this once a month".
- What goes in. The fields, in the order you would fill them.
- What comes out. The number, list or decision you actually need, and where it sits on the screen.
- What must never break. Rounding, currency, dates, empty states.
Then add one line about data. On a paid plan, ask it to save entries with personal storage so each person who opens it keeps their own data. On the free plan, ask for export and import buttons so your data lives in a file you control.
Five builds worth trying
A project quote calculator
Build a quote calculator artifact for freelance design projects.
I add line items with a description, hours and an hourly rate, or a fixed price.
Show subtotal, a discount field, tax at a rate I set, and the total in large type.
Add a "buffer" percentage that is included in the total but hidden from
the client view. Include a clean client view I can screenshot and send.
Save my rates and last quote with personal storage.
A meeting cost timer
Build a meeting cost timer. I enter the number of people and an average
hourly cost, press start, and it shows the running cost in large numbers
updating every second. Show elapsed time underneath. Add a pause button
and a summary line I can copy when the meeting ends.
It looks like a silly little tool. Put it on a shared screen once and see how the meeting changes.
A decision matrix
Build a weighted decision matrix. I add options as columns and criteria
as rows, give each criterion a weight from 1 to 5, and score each option
from 1 to 10. Show the weighted total per option, highlight the winner,
and show how much each criterion contributed. If two options are within
5 percent, say it is too close to call.
A client onboarding checklist
Build an onboarding checklist for new clients. Steps are grouped into
Contract, Access, Kickoff and First delivery. Each step has a checkbox,
an owner (me or client) and an optional note. Show progress per group and
overall. Let me reset the list for a new client and keep a list of past
clients with their completion date. Use personal storage.
A reading log with notes
Build a reading log. Each book has a title, author, status (to read,
reading, done), start and finish dates, a rating and a notes field.
Show books finished this year, average days per book, and a simple chart
by month. Let me search my notes. Add export to JSON and import from JSON.
Iterate in small steps
The first version will be close, and rarely exactly right. Fix it with short, specific requests instead of rewriting the prompt: "move the total above the line items", "make the dark mode colors match the light mode ones", "the tax field accepts letters, fix that". In documents you can highlight text and use Edit with Claude to change exactly that part.
If the artifact shows an error, use the Try fixing with Claude button next to it. And if you want to explore a different direction without losing the current one, edit an earlier message. That creates a new branch of the chat with its own artifacts.
Artifacts that think, and artifacts that connect
Two features turn a tool into something closer to an app.
First, an artifact can call Claude itself. A writing coach, a quiz that adapts, a tool that turns messy notes into a clean summary. The first time a new artifact wants to use Claude, it asks your permission. When you share it, people use their own Claude account, so their usage counts against their plan, not yours.
Second, on paid plans an artifact can use the apps you have connected to Claude, such as your calendar or project tool. The first time, Claude shows which tools it will use and asks you to approve them. Everyone who opens a shared artifact connects their own apps. Keep these read only until you trust them.
Share it carefully
Artifacts start private. Click Share and choose who has access. The sharing article lists what each plan allows.
Two things to check before you send a link. Look at whether the artifact uses personal or shared storage, because with shared storage everyone sees the same data. And never put anything sensitive in an artifact whose storage you did not choose yourself.
If you work in Claude Code, the same idea exists there: Claude can publish session output as an artifact, and /artifacts lists everything you have published.
Pick the tool you would otherwise build in a spreadsheet this week. Describe who uses it, what goes in, what comes out and what must never break. Then use it for a few days before you add a single feature.