Przejdź do treści
tl;dr of Claude Dashboards
A Claude dashboard is a live artifact, a page that Claude builds in Cowork that can fetch your current figures from sources when you open it. Available on paid Claude plans, on the desktop version of Claude.
Building it takes about five minutes. Simply tell Claude where the data is stored, describe the panels you want, and it creates it.
It does not update in the background. You have to open it for it to refresh the data.
I created one against a 400-row sales sheet to see if it was able to render figures correctly and give me a workable dashboard. It was able to read the entire file, including some traps I put in to test how deep it went. The charts needed more checking, though. Two were drawn wrong before Claude caught the problem itself.
A Claude dashboard is a live page that pulls your data into one customized view. These are what Anthropic calls live artifacts. Launched in April 2026, they are effectively a way of turning Claude into a reporting tool, and even a sales coaching dashboard.
To build one takes about five minutes. Point Claude to a data source, such as a spreadsheet or your CRM, describe the types of panels you would like included, and it writes a page that can go and fetch updated data whenever you open it.
The key thing here is that you need to open it. It doesn’t currently auto-populate, and if you close your laptop, the file just remains as it is.
I built one myself using a 400-row sales sheet, with every total calculated before Claude saw the data, then checked what it produced against the source.
What Is A Claude Dashboard?
A Claude dashboard is a live artifact. That is a persistent HTML page that Claude creates in Cowork, which reconnects to your selected data sources whenever you open it.
The difference between this and a regular Claude answer is where it lives and what it does when you come back to it again. If you ask Claude a question, you get an answer in the chat, whereas a live artifact is saved separately in the sidebar and keeps a version history you can scroll back through.
This is a feature that is limited to certain Claude account types and must be used with the desktop app, not the browser version of Claude. Live artifacts are available on paid Claude plans, such as Pro, Max, Team, and Enterprise. Many people are being guided to try it by Claude suggesting it in the message box.
If you already use Claude Projects, a live artifact is the output rather than the workspace. Projects hold the context.
Why Build a Claude Dashboard?
Every Monday morning, you query your data to check the current status. This could be a few clicks; it could also take hours of going across different data sources.
The data is there, stuck in a CRM that has a dashboard designed by the vendor’s team, but doesn’t necessarily have what you need for your daily work.
A Claude dashboard allows you to describe the panels you want in just a few sentences, and your page is built to your specifications, rather than a fixed template. For example, you are a sales manager and want to know each week which deals have stalled, what the pipeline looks like by stage, and which of your reps are behind on their follow-ups. Three panels, a sentence each, with relevant data.
It is also quicker. Your engineering team could build you an all-singing, all-dancing dashboard, but by the time it ships the number you were chasing has changed. A Claude dashboard you can adjust yourself.
Three Things Are Called Artifacts
The word artifact actually has three separate meanings in Claude.
Chat Artifacts
Launched in 2024, these are the output panel that opens beside a normal conversation in Claude.
Cowork Live Artifacts
Launched in April 2026 and are available on desktop, paid-for plans only. This is the only type of artifact that can go back and access data sources on its own.
Claude Code Artifacts
Launched in June 2026, they are pages generated from a coding session, created from the context of that specific session, including your codebase.
| Chat artifacts | Cowork live artifacts | Claude Code artifacts | |
|---|---|---|---|
| Launched | 2024, MCP added October 2025 | 20 April 2026 | 18 June 2026 |
| Where it runs | Web, desktop, mobile | Claude Desktop only | Any browser |
| Plans | Free upward | Pro, Max, Team, Enterprise | Team and Enterprise beta |
| Refresh | Static unless rebuilt | On open, with a short cache | Updates during the session |
| Udostępnianie | Public link available | Team and Enterprise, internal only | Shareable link |
One other thing to note is that Scheduled Cowork tasks run in the cloud on a schedule that you set, with no device online. This is genuinely automatic, whereas a live artifact is not.
What Can Claude Build a Dashboard From?
There are three kinds of sources that you can build a Claude dashboard from, and the ones you select dictate what the dashboard can and cannot do.
A CRM System
Pipedrive, Salesforce, HubSpot. The connector authenticates and holds the credentials itself. So nothing about your data has to be made publicly readable.
HubSpot has a first-party connector, but a Super Admin or user with App Marketplace permissions must connect it before anyone else can. It works with HubSpot’s free plan, but still needs a paid Claude account. Pipedrive launched its own MCP server in June 2026 on all plans, added as a custom connector. Salesforce uses its own hosted MCP server with OAuth and an external client app, so it takes longer to configure.
All three inherit the permissions set up in your CRM, so someone without full access will see less pulled into their dashboard. Good for privacy, awkward when two people compare figures.
Spreadsheet
If you are a bit old school, then a spreadsheet that is regularly updated can also be used as a data source. Google Sheets, or a CSV file in Drive. Point Claude at the file and it can read it.
The good news is that it reads all of it, in one go, with no pagination to handle. I tested this with a 400-row spreadsheet and it read the lot.
There is a catch, though. Permissions. A dashboard is effectively a web page, and it cannot log into a Google account and doesn’t have its own Google login. So it can only read a sheet that it doesn’t need to log into. To get a live refresh, you need to set the sheet to “anyone with the link can view”, which opens up issues of security and data privacy. The sheet that I used was fabricated, with made-up data and deals, but for sensitive data this isn’t the kind of information you want anybody with the link to have access to.
Warehouse or BI layer
For the more technically minded, Snowflake, BigQuery, or a Power BI semantic model can be used. There is no on/off toggle. You point Claude at a custom MCP server, which someone has to run and maintain, and it has to be reachable over the public internet because Claude connects from Anthropic’s cloud rather than your machine.
One of the earliest builds was from Kurt Buhler in April 2026, who connected a live artifact to a Power BI semantic model and queried it in DAX. He described the results as slow but interesting, and thought the slowness could be fixed.
If your data already lives in a warehouse, you likely have a BI tool that can do scheduled refreshes and shareable links, neither of which a live artifact can do. So while it theoretically can be done, live artifacts are probably a step behind what you already have in this instance.
How To Build a Claude Dashboard
I attempted to build my own Claude Dashboard from scratch, using fabricated sales pipeline data, to see how easy it was to create. It took a few minutes of work, a bit of waiting, and produced something quite insightful.
Before you start
You need Claude Desktop running the latest version, a paid Claude plan (I used Max), and a data source with a connector attached (in this example, Google Drive).
I would recommend connecting data sources before you create the artifact, as while often Claude will prompt you to connect a source mid-build, by having it attached it can get straight to the data rather than stopping to ask you.
The steps
1. First, get the data somewhere Claude can reach
A sheet in Google Drive, or a CRM with its connector switched on. Keep a spreadsheet to a single tab. The Drive connector has been documented as reading only the first tab of a multi-tab file, without flagging that it had ignored the rest.
2. Connect the source
Settings, then Connectors. On Team and Enterprise plans, an Owner has to add custom connectors at the organization level before anyone else can connect to them.
3. Select Cowork, not Chat
There is a small selector in the bottom left of the message box. Pick Cowork. Live artifacts do not exist in a normal chat.
4. Write the prompt
Name the source, name the figures you want, and say it should refresh when you open it. Mine:
Build me a live sales dashboard from the Google Sheet called sales-pipeline-400. Show total pipeline value, deal count and value by stage, a leaderboard of deals and value by owner, and deals with no contact in the last 21 days. Refresh from the sheet each time I open it.
If your data is in a CRM, Claude will typically work out what connector makes the most sense. For example, if HubSpot is connected, it will work out that that’s the place that the data is located. Here is a prompt to use that pre-existing CRM:
Build a live sales dashboard that stays up to date. Pull my team’s pipeline data and show open deals by stage, deals stalling with no contact in 21+ days, this quarter’s close rate versus last quarter, and a leaderboard of reps by calls logged this week. Refresh it each time I open it so I’m always looking at today’s numbers, not a snapshot from whenever I last asked.
5. Answer the questions before it builds
Claude came back with three questions before writing any code. How should the refresh work, given the sheet is private. Should closed deals count in the totals. And one row looked like a placeholder at 27% of total value, so what should happen to it.
The first question is the one that matters, and the next section explains why.
6. Watch what it does without being asked
Claude loaded a dataviz skill on its own and read two of its reference files, so the colors and chart types in the finished dashboard came partly from a skill I never invoked. It wrote itself a five-step task list before starting. And during verification, it caught two of its own rendering bugs, plus a currency error from earlier in the conversation, and then fixed all three before handing anything over. It checked its own output against the underlying numbers rather than assuming the charts were right.7. Find it again
The artifact saves to the Artifacts view in the sidebar. Open it from there and it fetches your current figures.Do You Need To Publish Your Data To Build a Claude Dashboard?
From a file, yes. From a connector, no.
A live artifact built from a private Google Sheet cannot read that sheet until the sheet is readable by anyone with the link.
Claude can’t work around this either. When I tested, it checked the file permissions, found me as the owner, then read the schema of the Drive sharing tool and found it can only grant permissions to named email addresses. There was no anyone-with-the-link option available to it, so it stopped the process and asked me to do it manually.
The best way around this is to only pull data from a place, such as a CRM, with a connector.
Do Claude’s Dashboards Update Themselves?
Not on their own. The dashboards update when you open them. They do not update in the background while they are dormant. What you have is a page that knows how to fetch data when you ask it to. Close your computer and nothing is happening. Anthropic’s own documentation says that opening the artifact pulls from your connected apps, with a short cache holding recent data so the page can load quickly. There is a refresh button that you can press to force a new query. There is no background execution. A feature that does run on its own is scheduled Cowork tasks; these run in the cloud with no device online. This is not what live artifacts are. Also, if you work across multiple machines, your dashboard works locally. Switch from your desktop to your laptop for a trip and your live artifacts will not be there. One thing to note from my build is that if the fetch fails, for whatever reason, there is a read-out of that at the bottom of the artifact. The status line states whether the data is live or is from a stored copy, offering a failsafe so you don’t assume the figures are live.Can You Share a Claude Dashboard?
Only on Team and Enterprise plans, and only inside your own organization. Pro and Max, while they can create them, cannot share or publish live artifacts at all. No plan can make them public.
If you were planning on sending a client a dashboard link, this feature will not work.
Sharing inside an organization also has some limitations. There are no external links, and no picking individual recipients to share your view. What appears in a shared dashboard also depends on the viewer’s own access, so the same page can show different figures to different people.
Are Claude’s Dashboards Accurate?
Yes, the totals were right. The charts needed checking.
How I tested it
I created two spreadsheets. One with 20 rows of invented pipeline data, and a second more substantial 400-row sheet. Each total was worked out and written down before I handed Claude either file.
On the larger 400-row file, I created a few things to see how accurately and deeply the read went. There were five larger deals planted deeper into the file, past row 100. If Claude only read the first chunk of the sheet, it would miss these and the numbers would come up short. The theory I was testing was whether it would read the file in full or switch to search after the first hundred rows.
I also put in a second test, one deal stage that only appeared once in the entire file on the very last row. If the dashboard pulled out seven deal stages, rather than the six that appeared in the first 399 rows, then I would be able to ascertain that it read the file in its entirety.
The totals
Every figure was correct. All five of the planted deals were present and the full seven deal stages were pulled in. Each owner total was correct as well. It was able to read the entire file and report accurately.
The charts
On a first attempt, the charts weren’t reliable, although Claude did flag this before it delivered the dashboard.
Claude drew a bar chart and worked out how wide it should be by comparing the top value against the top of the axis. But it set the top of the axis too low, lower than the biggest number in the data. The result was that the bars were larger than the chart they were in.
Second, one bar came out at 99.766% of the chart’s width, so it looked like the largest section. The issue was that it was only 29% of the actual total. It looked plausible but was actually incorrect.
The automated layout check it ran reported everything as fine. The problems were only visible when it printed the actual bar widths and compared them against the scale. So if you are taking this into a meeting, run a quick check against your source beforehand. The totals held up across the entire file. The chart drawing them did not.
Should You Build a Claude Dashboard?
Yes, if you would like a view of your own figures and data without raising a ticket or learning a BI tool, and you can live with it updating when you open it rather than constantly.
Not yet if this is something you want to send to a client, if it needs to be current without anybody looking at it, or if it cannot sit behind a link that works for anybody who has the URL.
It’s worth thinking about what else you can connect it to. A lot of what a sales manager wants to see- who has been spoken to, what’s been agreed- doesn’t always live purely in the CRM system, but in calls that have been made.
It’s worth thinking about what else you can connect it to as well. A lot of what a sales manager wants to see- who has been spoken to, what’s been agreed- doesn’t always live purely in the CRM system, but in calls that have been made. tl;dv’s MCP server is read-only and OAuth-authenticated, so it connects to Claude without an export to publish.
Whatever you connect, whatever dashboards you create, just make sure to spot-check against the source before acting on it.
FAQs About Claude Dashboards
How do you build a dashboard in Claude?
Open Claude Desktop, select Cowork in the message box, and connect your data source. Then describe the panels you want and name the source. Claude builds a live artifact that saves to the Artifacts view in the sidebar. You need a paid Claude plan, and it only works on desktop.
Do Claude dashboards update automatically?
Not on their own. They fetch fresh data when you open them, with a short cache and a refresh button in the header if you want to force it. Nothing happens while the dashboard is closed. Scheduled Cowork tasks are a separate feature that does run in the cloud without your machine.
Can you share a Claude dashboard?
Only on Team and Enterprise plans, and only inside your own organization. Pro and Max can build them but cannot share or publish them, and no plan can make one public. Anyone you share with sees data through their own connector permissions, so their version may show less than yours.
Does Claude read every row of a large spreadsheet?
In our test, yes. All 400 rows were read and every total matched the source, including deals planted deep in the file to catch a partial read. Larger files, and CRM sources that hand over data in pages, were not tested here.
Does your data have to be public to build a Claude dashboard?
From a Google Sheet, yes. The sheet has to be viewable by anyone with the link, because the dashboard has no login of its own. A CRM connector avoids this, since it authenticates on its own behalf and nothing gets published.
