What are Claude Dashboards? Analyzing and visualizing data with Anthropic's Claude.ai

Key takeaways

• You can create Dashboards (in beta) in Claude Desktop or claude.ai: These are live artifacts that can query data sources via Claude connectors with MCP servers. They can use extracts or live queries that can stay up-to-date, and you can share the artifact dashboards with other people.
• Solid data foundations are more essential now: Claude Dashboards are a new front-end option available to you. The quality of your semantic models and how they are governed decide how fast you can adopt new front-ends.
• A remote MCP server is required for a live connection: Currently, Microsoft's Fabric IQ MCP server doesn't work for this use-case and its Power BI MCP servers (preview) are untested.
• Claude Dashboards are different from Fabric Apps: Claude Dashboards provide a simpler, more constrained experience which focuses more on analyzing and visualizing data. While Claude Dashboards might be a simple, lean option for individuals or small teams (i.e. personal BI scenarios) there are clear and present challenges for data governance and compliance that likely make this a difficult option to consider for enterprise scenarios.

This summary is produced by the author, and not by AI.

Create and share a dashboard from Claude

Anthropic released Claude Dashboards in beta on 8 October 2026, together with Claude Motion. Both are beta tools, not ready for production yet, but clearly evidence of things to come. This lets Claude use connectors to query data and use that data to build web dashboards as live artifacts that you can view, interact with, and share. You can thus ask Claude to create a dashboard on data from Microsoft Fabric or Databricks, such as a Power BI semantic model, or a Fabric data warehouse or lakehouse. The following is a short demonstration of what this looks like:

This might seem similar to Fabric Apps, but there are some key differences, which we'll discuss later. In this article, we briefly introduce Claude Dashboards and Motion, focusing on the dashboarding and data analysis. We'll discuss our first impressions and opinions and our view of where this might fit, and what some of the current challenges and considerations are.

What is a Claude dashboard?

An artifact is, in Anthropic's own words, "anything Claude makes for you that you'd put in front of someone", such as "a design, a deck, a document, a dashboard, or a small interactive tool." Artifacts are by default private to the user, but their purpose is to be shared with people in your organization, or, if permitted by your organization's admin, anyone with the link to the artifact.

Claude Dashboards are a type of artifact. As with other artifacts, you ask Claude to build and change a dashboard from the chat, whether in your browser, the Claude Desktop app or the mobile app. Each Claude Dashboard has two parts: one HTML page and one or more datasets. The page is what people look at and click on; the datasets are listed in the Sources pane on the side. A dataset can be a local file extract or a live connection to a data source.

The diagram below shows how the pieces fit together. You build the dashboard by chatting with Claude in Claude Desktop. Claude uses a connector to query Fabric data, such as a semantic model or a lakehouse or warehouse. A local connector on your own machine can only feed extracts into the dashboard, while a remote connector in the cloud lets the dashboard rerun live queries. Each query runs with the user's own identity. You can then share the dashboard with colleagues, who open it as themselves.

What do Claude Dashboards mean for semantic models and BI?

It's clear that data analytics and BI is changing, fast. However, amidst all that change, some things are only becoming more important: data quality, governance, and your semantic model. More and more front-end options are available to you, with Claude Dashboards just being the latest one. At Tabular Editor we're striving to ensure that irrespective of where you go or how you consume your data your semantic models will be quality, performant, and trustworthy, both with and without the use of AI.

How do Claude Dashboards work?

Let's now take a look under the hood. We update our articles regularly, but this is a brand new feature that will evolve fast. Check the update date at the top of this article and re-verify before deciding big things based off it.

How do you connect your data?

To use Claude Dashboards with your own data, you need a connector. Connectors are MCP servers that you either provide yourself (custom connectors) or which are provided by a first party (Anthropic, such as its Microsoft 365 connector) or a third party (the service's own vendor, such as Atlassian). As of October 2026, Microsoft provides no ready-made connector for Claude; you can add its Power BI MCP servers (preview) as a custom connector after registering an Entra app. There are connectors available for Databricks, Snowflake, and other data sources. A custom connector or MCP server that you already have connected can be used to query a data source and construct the dashboard; however, to have live queries that refresh, this connector must be hosted remotely. For example, the Tabular Editor 3 MCP server runs on your own machine: it can supply data while Claude builds the dashboard, but the published dashboard can't call the TE3 MCP server on your machine.

Much like data connectors in Power BI, they range from vetted, vendor-built connectors to custom ones that anyone can build and host:

There are thus two ways to make a Claude Dashboard:

• Using a connector, Claude queries and extracts the data as CSV and JSON files to build a dashboard. To refresh it, it must re-take the extracts. Anyone who can open the dashboard can read these files.
• Using a connector with a remote MCP server, Claude links queries to visuals and data points to build a dashboard. The queries can be re-executed to refresh the dashboard, and each query runs with the viewer's own connection and permissions.

Is a dashboard live or an extract?

Unfortunately, Claude Dashboards currently do not provide a clear way to distinguish between a local or live dashboard from the user interface. For this, you need to click the database icon ('see where the numbers come from') and then inspect the 'query'. If you don't see a query but rather a named file, then it's working with an extract. This is clearly a governance challenge that can hopefully be addressed with future improvements to tools and context.

How do you create and share the dashboard?

Once you have a connector set up, you just have to prompt Claude to create the dashboard. By default, the experience seems to be less focused on making beautiful, interactive dashboards, and more about answering specific questions or problems about the data. As such, Claude focuses more on creating visuals that augment a particular analysis and try to tackle the question and its underlying explainers.

Once you are satisfied with the result, you can share the artifact. This uses the same artifact sharing system that already exists in Claude today, and has the same governance or distribution controls. Admins of Team and Enterprise plans have some controls and tools to manage this, but there's a clear governance challenge. For more information, see Get started with Claude Dashboards, the artifacts admin guide for Team and Enterprise plans, and how connector calls work for viewers.

Next to Share, there is an Export tab. It does not let you download the dashboard itself as files; it only sends the dashboard to a connected destination: Amplitude, Grafana Cloud, Hex, Mixpanel, Omni Analytics, Perplexity Computer, PostHog and Sigma, with monday.com marked as coming soon. As of October 2026, Power BI and Microsoft Fabric are not on that list. The only way to get data out yourself is to download a source's rows as CSV from the Sources pane.

People you share the dashboard with can also leave comments on it:

Where do the numbers come from?

A dataset is a single table with a short title, an optional one-sentence description, and its source. A single dataset is capped at 8 MB, so row-level facts rarely fit and you aggregate first; a live connector does that aggregation on the platform, so the cap rarely matters. Each dashboard also has its own storage, which at the time of writing shows a limit of 100 MB for its database and 1 GB for files.

Every value on the page must come from a dataset. This "dataset" is obtained from the queries (for a live source) or files (for extracts). In the Sources pane you can click a number to see its query, its rows and how fresh it is, and download the rows as CSV. Calculations done in the page itself, such as a share of total, show as a d3/JavaScript formula rather than a query; they never touch the semantic model. Editors can also edit, test and restore a live query from the Sources pane.

The refresh button at the top reruns the live queries, and hovering over it shows how fresh the data is.

You can also click a number and choose "Ask Claude" to have Claude explain how it was calculated:

A flexible front-end

Since the page is HTML, there's a lot of flexibility in what you can build (charts use d3.js). Custom controls like tabs for multiple pages, cross-filters, slicers for date and text, searchable tables, and even scroll-driven pages where a chart stays in view are all possible. There are some limits, though. The following common web development tools and browser features are not supported:

• network access from the page (i.e. you can't call a web API or load anything at view time, such as fonts or map shapes; if you want those, they must be coded into the page)
• external libraries (except d3)
• React or JavaScript modules
• storage or cookies, iframes or forms
• files over 256 KB in the dashboard's own store, which includes the page itself

How does a Claude dashboard compare with a Fabric app?

For Microsoft data professionals, there are some obvious comparisons to be made between the recently released Fabric Apps and Claude Dashboards.

Building an app vs building a dashboard

In our Fabric Apps article we said "if a Power BI report is LEGO, then a data app is a 3D printer". Extending this, perhaps Claude Dashboards are something in the middle; a 3D printer that's constrained to certain options, and tailored toward answering data problems more than building anything that you want. For instance, Claude Dashboards don't have the same possibilities for integration with your data platform, including things like writeback. Indeed, Claude Dashboards seem simpler and leaner compared to Fabric Apps, since they don't require the user to create a code project, use a CLI, or deploy. They add certain guardrails and constraints via the connectors and artifacts infrastructure and undoubtedly some context that Anthropic provides, so the entire experience from agent to tooling is tailored toward getting a good result.

That simplicity has a cost, though: you can't view or download the project code behind a Claude dashboard in the interface, so the dashboard lives only in Claude (unless you export it to one of the supported services). With a Fabric App, you start with the code and keep it in your own repository, although a visual editing experience for Fabric Apps is also coming.

This also means that, as far as we can tell, you can't put a Claude dashboard in source control. Claude does keep some version history: each publish of an artifact becomes a version, and you can choose which version viewers see. Admins on Enterprise plans can also list artifacts and retrieve a version's content through the Compliance API. However, there is no way in the interface to download the dashboard's HTML, scripts and datasets as a project, so you can't review changes in a pull request, run CI/CD, or deploy the same dashboard elsewhere. For anything beyond personal or team analysis, this is a significant limitation.

Furthermore, Claude Dashboards do seem to be lined up for deeper integration with the Claude for Business tooling, with Motion being the example given by Claude. Anthropic's launch video shows a good example of this.

Governance, security, and distribution

When people are creating BI artifacts and sharing them in an organization, it's important that they're doing this from a centralized source of truth. This ensures that everybody's speaking the same language, and it's one of the most basic yet challenging parts of self-service business intelligence.

It's also important to have an overview of who is creating and distributing BI content in an organization and what data sources they're using. This is why governance is only becoming more important with AI, as it's getting easier and faster to create analytics and dashboards in Power BI and Fabric. This has been a struggle since the beginning of Power BI, and many options have come into place to facilitate this.

Clearly, this is an area where Claude Dashboards differ a lot from anything in a data platform. Claude Dashboards have limited options for governance tied to the artifact system itself, and there doesn't seem to be a way to easily discover and reuse dashboards from a separate gallery (though maybe that's coming). For what the creation experience carries in convenience and potential, at an organizational level this spells a potential nightmare for governance and administration. These challenges stretch from the dashboards to the underlying data, as it's likely that Claude will re-calculate metrics in the dashboard visuals rather than using what's available in a semantic model. It also could ignore important organizational context or guardrails set up in a data platform.

These things might be less of a challenge for smaller organizations or personal BI scenarios. However for the enterprise it presents some pretty daunting and hairy issues about how to avoid incidents or inefficiency with using data. Nonetheless, it's clear that as more users and organizations adopt tools like Claude or ChatGPT, the convenience of having your data analysis at hand without much technical complexity will create pressure and appeal for self-service BI users.

Where Claude Dashboards fit, today

Claude Dashboards are born from chats and built by agents. This means they can be whipped up in minutes and lend themselves well to answer ad-hoc questions and fast prototyping. Today, they lack important features that longevity demands: source control, fine-grained permissions on extracts (such as row-level security) and sharing (such as group-based access), reusable components, and so on.

Every design choice is made again in every dashboard. In this one, for example, we added filters to table headers, but there is currently no way for other dashboards to import this component as a reusable pattern. Wherever this pattern is needed, it may be implemented similarly but differently, leading to a UX kaleidoscope across dashboards.

The page of a Claude dashboard can currently only be changed by Claude itself. Fabric Apps, in contrast, are code solutions you fully own and natively run on the data landscape you spent time and effort getting right. The ease with which Claude Dashboards are created might tempt some to go back to the days of read-all extracts and encoding business logic in de novo formulas that duplicate the governed ones. Live queries against a semantic model avoid the issue altogether, because they use the model's measures and run with each viewer's own permissions. The Sources pane will show for any number where it comes from: a live query, a file, or a formula in the page. That's a good start for data lineage, right where you need it.

So where do we see them fitting today? If you use Claude, we think the purpose and envisioned longevity of the dashboard are a useful rubric:

• Ad-hoc exploration and prototyping: When you have a question about data while working with Claude, a new dashboard is just a prompt and a few minutes away. As we said, this is very useful for prototyping a dashboard on real, governed data before it's built as a governable Fabric App. A prototype like this can also reveal gaps in the model, such as a measure that's missing. With the Tabular Editor CLI, you can fill those with a few commands:

# 1. Look for a delivery cost share measure in the model
te find "Delivery Cost %" --in all -s "<workspace>" -d "Invoiced Sales"

# 2. Add the missing measures to the live model (no deploy, no refresh)
te add "Invoiced Sales/Delivery Cost" -t Measure \
  -p "Expression=SUMX('Budget Rate', CALCULATE(SUM('Invoiced Sales'[Delivery Cost])) / 'Budget Rate'[Rate])" \
  -p "DisplayFolder=Measures" \
  -p "Description=Delivery cost in EUR, converted at the budget rate." \
  --save -s "<workspace>" -d "Invoiced Sales"
te add "Invoiced Sales/Delivery Cost %" -t Measure \
  -p "Expression=DIVIDE([Delivery Cost], [Revenue])" \
  -p "DisplayFolder=Measures" \
  -p "Description=Delivery cost as a share of revenue." \
  --save -s "<workspace>" -d "Invoiced Sales"
Changes saved to server.

# 3. Query the new measure by system
te query "EVALUATE SUMMARIZECOLUMNS(Region[System], \"Delivery Cost %\", [Delivery Cost %])" \
  -s "<workspace>" -d "Invoiced Sales"

• Visual explainers: The page is HTML so you can show whatever code can specify. This makes Claude Dashboards useful for a quick analysis to share with the team, or, with Motion, a presentation that tells a polished narrative.
• Dashboards with broad audiences: When a dashboard has to serve many people for a long time, build it as a Power BI report or a Fabric App. Use the Claude dashboard as the prototype, then build the real thing where your source control, deployment and governance are already in place.

Before going all-in on Claude Dashboards, consider the gaps around data retention, governance and security that are there, at the time of writing.

Further Reading

• Fabric Apps explained: Visualization as code in a data app dashboard (Tabular Editor). Our introduction to Fabric Apps, the code-first option this article compares Claude Dashboards with.
• Get started with Claude Dashboards (Anthropic). Anthropic's guide to building and sharing dashboards, and the place to check for changes while the feature is in beta.
• Register Power BI MCP servers for MCP clients (Microsoft Learn). The steps to register an Entra app so that Claude can use Microsoft's hosted Power BI MCP servers as a custom connector.
• Get started with the Fabric IQ MCP server (Microsoft Learn). Microsoft's remote MCP server for finding and querying Power BI content, which at the time of writing does not work with Claude Dashboards.

Conclusion

Claude Dashboards provide a way to create re-usable data analyses and visualizations from Claude instead of a BI tool. While many teams and individuals were already exploring this, it provides special tools and context to help you analyze your data in Claude. However, there are clear and present challenges with data governance, security, privacy, and more. These challenges will only be exacerbated as AI adoption advances in the enterprise.

As more users and organizations adopt tools like Claude, having data analysis at hand without much technical complexity will put pressure on self-service BI. The quality of your semantic models and how they are governed will decide how fast you can adopt new front-ends like this one.