How to build a live business dashboard in Claude
MarkdownConnectors, Cowork, and published artifacts: a ready-to-use dashboard prompt, metrics, limitations, and read-only access.
Read full article
Key takeaway
In July 2026, Claude learned to connect to your data and build a business dashboard - a live metrics panel that updates itself every time you open it. I show you how to put together an overview of revenue, leads, and pipeline from your spreadsheets and CRM in one evening - without an analyst, an expensive analytics subscription, or hosting. And I’ll be honest about where this thing falls short.
Ailoot documentation describes how to connect models and work with the API.
What is a live dashboard, and how is it different from a regular spreadsheet?
Key takeaway
A live dashboard is a page Claude builds from your data that automatically pulls fresh figures from connected services every time you open it. A regular spreadsheet freezes at the moment it’s exported. A live panel shows what’s happening in your business right now, as you look at it.
It looks like this: you open one page and see monthly revenue, new leads, and how many deals are in your pipeline. You don’t export again, copy things into a spreadsheet, or consolidate numbers by hand. You just open it - and the figures are already fresh.
Previously, an artifact in Claude (a built page with a chart or calculator) was a snapshot. The data was baked in at the moment Claude built it. Open it a week later - the figures were still the same. Now such a page can reach into your Google spreadsheet, CRM, payment processor, and show what’s current. Anthropic describes it plainly in its documentation:
An artifact can call MCP connectors every time someone opens it, so the page shows up-to-date data instead of a snapshot from the session in which it was built.
The key word is connector. It’s a ready-made connection between Claude and an external service: Google Sheets, Notion, Stripe, email. You grant access once, and after that the page gets data from there on its own. It works on top of the open MCP protocol - if you want to dig deeper, I have an explainer on MCP servers for Claude.
Why does a report you put together by hand go stale by morning?
Key takeaway
A manual report goes stale because the business keeps moving, and the spreadsheet doesn’t. While you were consolidating yesterday’s revenue, new payments and leads came in. By morning, the numbers are already different. So you either look at an outdated picture, or spend an hour every morning rebuilding it. A live panel eliminates that work entirely.
It’s a familiar story. Once a week, you open five tabs: your bank, payment processor, CRM, advertising spreadsheet, and something else. You copy numbers into one file, add color, calculate the difference from last week. Two hours later, you have a beautiful report. And a day later, it’s dead - the data has already changed.
The point of a live dashboard is simple: you do the work of “turning data into a picture” once, with a prompt. After that, the picture updates without you.
What happened in 2026: Claude learned to connect to data
Key takeaway
On April 20, 2026, Anthropic launched live artifacts in the Cowork app - personal panels that pull in your data. Then, on July 16-17, 2026, published artifacts in Claude Code gained the same ability: the page calls connectors every time it’s opened. From that moment, an artifact stopped being a picture and became a mini-tool that shows live figures.
It’s important not to get confused here. Claude now has two similar surfaces, and both provide a “live panel that updates itself.” I discuss their differences and how to choose between them separately. First, here’s what happened.
Before spring 2026, every artifact was static. Claude built a page and permanently embedded the data in it. In April, live panels appeared in the Cowork desktop app. In mid-July, connectors were also enabled for published artifacts - the ones you build in Claude Code or chat and publish to your private address. The July 17 release note puts it this way:
A published artifact can now call MCP connectors every time someone opens it, so the dashboard shows live data and can perform actions on request instead of showing a snapshot from the session in which it was built.
For an entrepreneur, in plain language, that means the metrics panel you put together in an evening no longer goes stale a day later. It looks at your data every time you open the tab.
The business data you’ve gathered in one place becomes a picture in Claude that you can look at every day. CLAUDE.md and the business/ folder structure in a project work on the same principle, but for code - I covered that in my guide to a second brain in Claude Code.
Cowork or a published artifact: where should you build the panel?
Key takeaway
For a solo entrepreneur who checks their own figures, a panel in Cowork is simpler - it lives in the desktop app and pulls your connectors without asking unnecessary questions. A published artifact is for when you need to show the panel to your team: each person then sees data through their own access. Start with Cowork if the figures are yours and for your eyes only.
There are two surfaces, and they really are different. Here’s when to use each.
| What you need | What to use | How it works |
|---|---|---|
| Check your own figures | Live panel in Cowork | Lives in the app’s sidebar on your computer and automatically re-queries your connected services when opened |
| Show the panel to your team or a partner | Published artifact with connectors | You publish it to a private claude.ai address; each viewer sees data through their own access |
In Cowork, the panel is more seamless - it doesn’t keep asking you for permission at every little step. Anthropic says:
Most of the time, you won’t need to refresh manually: a short cache holds recent data so the artifact loads quickly, and it automatically re-queries your connected apps.
Published artifacts work differently, and it matters for trust: data is shown through the access of whoever opened the page. Your access has nothing to do with it. Two people opening the same panel will see different things - each only what they have access to. That’s right for a team dashboard, but unnecessary complexity for personal use. So my advice is: if the panel is just for you, start with Cowork.
What do you need to build your own panel?
Key takeaway
You’ll need three things: a paid Claude plan (Pro, Max, Team, or Enterprise - connectors don’t work on the free plan), data in one of the supported services (Google Sheets, Notion, CRM, payment processor), and 20-30 minutes for the first build. You don’t need code, a hired analyst, or your own hosting.
Let’s break down the barrier to entry as it is. To build a live panel, you need:
- A paid Claude plan. Connectors to services work on Pro, Max, Team, and Enterprise plans. There are no live connections on the free plan - you’ll get a static snapshot. Published artifacts require Claude Code version 2.1.209 or later.
- Data in a service you can connect. Google Sheets, Notion, Stripe, HubSpot, email, Slack - somewhere your numbers already live. If revenue exists only in your head and a notebook, first move it at least into a spreadsheet.
- Half an hour and a clear head. It’s important to understand which three or four numbers you really need to see every day. No technical skills are required.
You don’t need to pay separately for connectors - they’re included in the subscription. The connection directory at claude.ai/directory has grown to several hundred services, and the count grows every week.
How to connect your data: spreadsheets, CRM, payments
Key takeaway
Connecting takes a couple of minutes and goes through the service’s standard sign-in screen. Open connector settings, find the service you need, click “Connect,” sign in, and confirm what you’re granting access to. From then on, Claude can get data from it for the panel. No code and no manually entering keys.
Here’s the process:
- Open connector settings. In the app or on claude.ai - go to Settings, then Connectors. The
claude.ai/directorycatalog of available connections is there too. - Find your service and click “Connect.” Google Drive and Google Sheets, Notion, Stripe, HubSpot, Gmail, Slack - choose where your data lives.
- Sign in to the service. The familiar sign-in screen opens (the same OAuth flow as when you sign in to another app with Google). It shows exactly what you’re granting access to. Read it carefully - the permissions you give the connector directly affect the security of your data.
- Done. The connector is connected. Now Claude can access this service when building and updating the panel.
One caveat about Google Sheets: the standard connector is usually read-only. That means the panel can read your sheet and show figures, but it can’t write back to the spreadsheet. Read access is more than enough for a metrics panel.
How can you build a metrics panel with one prompt?
Key takeaway
You build the panel with ordinary text: describe to Claude which numbers you want to see, which connector to get them from, and how to arrange them in blocks. The more specific the prompt, the less you’ll need to edit afterward. Name the metrics, period, breakdown, and source. Then Claude draws the panel right in the chat, and you refine it through more conversation.
Specificity makes all the difference. A vague request like “show me my sales data” will return a generic table that tells you nothing. Name everything explicitly, as you would in a design brief.
Here’s a ready-made prompt you can copy and adapt:
Prompt
Build a live business metrics dashboard
Build a live dashboard artifact with my business metrics. Pull data through my connected connectors whenever the page is opened.
Sections:
Top row - 4 large tiles: monthly revenue, number of new leads, value of deals in progress, average order value. Under each number - percentage change versus last month, a green upward arrow or a red downward arrow.
Revenue chart by week for the last 8 weeks.
Open deals table from CRM: client, amount, stage, number of days open. Sort by amount.
Sources: get revenue and leads from [name your connector - Google Sheets / Stripe / CRM], deals - from [name your CRM].
Add a period filter at the top: this week / 4 weeks / quarter.
If a connector isn't connected, show a clear message in that section saying which service to connect.
Claude will draw the panel in the side of the chat. Then you edit it in words: “put the tiles in one row,” “change the chart to a bar chart,” “add a comparison with the previous quarter.” Each version is saved, and you can roll back to any of them.
What dashboards do entrepreneurs actually need?
Key takeaway
The most useful panels answer one specific question for the owner. Revenue pulse, lead funnel, a command center spanning different services, a weekly advertising summary. Don’t try to cram everything in at once. One panel - one decision you make from it. Then you’ll open it every day. A panel with ten metrics is usually opened once and forgotten.
Here’s what people actually build:
| Panel | What it shows | Where the data comes from |
|---|---|---|
| Revenue pulse | Revenue, average order value, churn, accounts receivable | Stripe, payment processor, Google Sheets |
| Lead funnel | Deals by stage, forecast, stalled deals | CRM (HubSpot, amoCRM through a spreadsheet) |
| Command center | Everything in one place: newsletter subscribers, payments from email, tasks for the week | Email, Notion, spreadsheets |
| Weekly advertising summary | Spend, leads, cost per lead by channel | Spreadsheet with an export from advertising accounts |
Practical advice: start with the question, “What one decision do I make based on this panel?” A typical mistake is cramming in everything you might want to see. The team opens it once, there’s too much, and they never return. One panel - one question.
Live panel, Power BI, or DataLens: which should you choose?
Key takeaway
Claude’s live panel wins on speed and ease of entry: you build it in minutes using ordinary text, without an analyst or data warehouse setup. Traditional analytics systems like Power BI or Yandex DataLens are better for large datasets, scheduled updates, and access management. Choose Claude for a quick personal overview. Choose an analytics system for company-wide reporting.
A comparison on the points that matter to entrepreneurs:
| Criterion | Claude live panel | Power BI / Yandex DataLens | Excel dashboard |
|---|---|---|---|
| Time to first panel | Minutes, using ordinary text | Hours or days; a data model is required | Hours, formulas and pivot tables |
| Analyst needed | No | Usually yes | Somewhat |
| Data is live when opened | Yes, through connectors | Yes, through a configured connection | No, manual import |
| Scheduled updates | No, only when opened | Yes | Partially |
| Public link | Not for connector-based artifacts | Yes | Limited |
| Hosting required | No | Yes, a service or license | No |
Here’s the answer. DataLens is stronger for heavy analytics and runs locally, but you have to configure data sources and build the panel by hand. Claude wins at something else: build an overview in 15 minutes using ordinary text and pull data directly from Stripe, Notion, or email without setting up a data warehouse. It’s a tool for a quick answer to a question here and now, without trying to replace an enterprise reporting system.
Where live dashboards fall short: honest limitations
Key takeaway
The key limitation: a panel with connectors can’t be shared by public link - on Pro and Max plans, it remains private and just for you. Also: there are no scheduled updates (only when you open it), no full server-side backend, a single page, and a size limit of 16 MB. This is a tool for personal and team overviews. You can’t use it to make a public report on a website.
I’ll be direct so you don’t imagine a panel and then hit a wall. Here’s where live dashboards fall short.
I’ll start with public links. A panel that accesses connectors can’t be opened through a public link on any plan. Anthropic states this directly:
An artifact that accesses connectors cannot be shared via a public link on any plan. On Pro and Max plans, where a public link is the only way to share, an artifact with connectors remains private and just for you.
The logic is clear: a public link has no signed-in viewer, so there’s no account whose access can be used to get the data, and it can’t safely hand your numbers to an anonymous internet user. The practical takeaway: on Pro and Max, a live panel is a personal tool. If you want to show it to your team, you need a Team or Enterprise plan.
The second limitation is that the panel has no full service behind it. Anthropic calls it what it is:
An artifact is a snapshot of work, a single self-contained page with no backend. It does not store data submitted through a form, and it does not verify who opened it.
The rest, more briefly:
- No scheduled updates. Data refreshes when you open the page or click the refresh button. The panel won’t send you a report at 9 a.m. on its own - you need a traditional analytics system for that.
- Storage and calculations are out of scope. The page shows data but doesn’t store information entered into a form. Do heavy calculations before Claude; the page only displays the result.
- One page, up to 16 MB. There are no multiple pages or separate sections within it.
How can you avoid giving Claude too much business data?
Key takeaway
The rule is simple: connect only what’s necessary and grant read-only access where possible. A connector with write permission inherits your full permissions in the service - if something goes wrong, it can be used to read, send, or delete. Don’t grant automatic permission for actions that write or delete. Connect financial services deliberately, only when the panel really needs them.
Connecting data to AI is about trust, and there’s something to learn here. Security specialists call indirect prompt injection the main risk of 2026: when an agent has access to your data, contact with untrusted content, and any path outward, a leak becomes almost inevitable. I cover this vector in more detail in my guide to how someone else’s repository can attack Claude Code.
What to do in practice:
- Connect only what this panel needs. Need revenue? Connect the payment processor. Don’t connect email and calendar “just in case.”
- Read-only only. Where possible, grant read-only access. A metrics panel shouldn’t write anything to your services.
- Require manual confirmation for write actions. If a connector can send messages or change records, don’t set it to “always allow.”
- Check what the panel shows. Connectors break, and the data structure in a service changes. In one case, a data feed stopped working, and the panel quietly displayed incorrect figures for four days. Compare key figures against the source once a week.
Attention
Read-only doesn’t mean “completely safe.” One prompt can pull data from email, chats, and documents all at once. That’s why minimizing connectors is not a formality, but a way to limit the damage if something goes wrong.
As for how the mechanics work, requests to Google, Notion, or Stripe are made by Anthropic infrastructure from its own servers. That means that after you sign in to claude.ai and confirm access to the service, the connections should work normally. This is a conclusion from analyzing the mechanics, not an official vendor guarantee - check it with your own set of services.
A Pro subscription costs about $20 per month.
Where should you start today?
Key takeaway
Start with one panel and one question. Choose the metric you check most often - usually revenue or new leads. Put the data in one spreadsheet, connect it, and ask Claude to build an overview using the ready-made prompt above. Don’t try to cover the whole business at once. One working panel that you open every day is more useful than ten beautiful ones that you opened once.
Specific steps for today:
- Decide which one metric is most important for you to see every day.
- Make sure its data is in one service - at least a Google spreadsheet.
- Connect that service through Claude’s settings (Settings, then Connectors).
- Copy the ready-made prompt, keep one or two sections for your metric, and build the panel.
- Open it tomorrow morning and check that the figures are fresh.
After that, add one block at a time as you figure out what you actually look at. Building the first project and getting it to launch is helped by adding working steps in sequence.
Sources
Continue working with AI on Ailoot.
More details
Questions about How to build a live business dashboard in Claude
Does a connected dashboard prove that today’s figures are fresh?
No. Open the panel again and compare its figures with the connected source. Account for the short cache and confirm the data period; merely connecting the service does not verify the displayed result.
Will a published dashboard show every teammate my data?
Each viewer uses their own connector access, so verify the panel with the intended viewers. For your own private metrics, the article recommends starting with Cowork and only the minimum necessary data access.