Tableau Data Apps beta lets Claude and ChatGPT build governed apps

1 hour ago

7 min read

Cover with the title Tableau Data Apps beta lets Claude and ChatGPT build governed apps on a teal data skyline. Data Studios original cover

Tableau has opened a beta of Tableau Data Apps, custom interactive web applications that are written in code, hosted by Tableau and powered by its query, connectivity and governance stack. The announcement, published on the Tableau blog on November 3, 2026 by product management directors Anushka Anand and Joe Constantino, describes two ways to build them: in Tableau Studio, where a builder can switch between AI prompting and drag-and-drop editing, or with any third-party AI agent connected through MCP, with Claude, ChatGPT and Codex named as examples.

The practical change for BI teams is that an app generated by an AI assistant no longer has to live on a separate web server with its own copy of the data and its own database credentials. According to Tableau, a data app keeps database credentials inside Tableau, applies object-level and row-level security to every user, reuses the semantics already defined in published data sources, and stays current through extract refreshes or live connections. The developer documentation limits the beta to Tableau Cloud.

The vendor has not published pricing, edition requirements or usage limits for the feature. Everything below comes from the announcement and from the data apps section of the Tableau Extensions API documentation, verified on November 4, 2026.

··········

WHAT A TABLEAU DATA APP IS.

The Extensions API documentation defines a data app as a purpose-built interactive application built with code to help users explore data and complete workflows. The app queries governed Tableau data sources at runtime through the Extensions API and a curated set of REST API methods. Tableau hosts the front-end files, so no separate web server is needed for them, but the documentation states that this does not provide a runtime for your own server-side code.

Technically, a data app is a packaged extension stored inside a workbook. The package bundles an extension manifest (a .trex file describing a Dashboard Extension or a Viz Extension), a web page with its JavaScript, CSS and images, and the Tableau Extensions API library. The whole set is embedded in a packaged workbook (.twbx) and published with it. A single workbook can contain more than one data app.

Tableau separates the concept from its existing extension types. A Dashboard Extension adds a custom web application to a dashboard zone, and a Viz Extension adds a custom visualization to a worksheet through the Marks card. A data app is the complete application experience built on top of those placements. The Tableau Embedding API remains the option when a company hosts its own application and embeds Tableau views inside it; data apps reverse that arrangement, because Tableau hosts the application code next to the content.

Three Extensions API methods carry the data work. queryAsync() retrieves data from the configured data source, readMetadata() returns the source's metadata, and fetch() makes REST API requests limited by the site's API access settings and by the viewer's permissions. Supported sources are published Tableau Cloud data sources, identified by their content URL.

··········

TABLEAU DATA APPS AVAILABILITY BY PLATFORM.

The announcement and the documentation describe availability with slightly different scope. The blog post says apps built in Tableau Studio publish to Tableau Cloud or Tableau Server, and apps built with third-party agents publish to Tableau Cloud, Tableau Server or Tableau Next. The developer documentation for the beta states that it is available only on Tableau Cloud, and the hosted Tableau MCP service is also a Tableau Cloud service; Tableau Server customers have to self-host the MCP server.

........

Item	What Tableau has published
Status	Beta
Announcement date	November 3, 2026
Platform in the beta docs	Tableau Cloud only
Publishing targets in the blog	Cloud, Server, Next
Build options	Tableau Studio or MCP agents
Agents named	Claude, ChatGPT, Codex
Hosted MCP endpoint	mcp.tableau.com (Cloud)
Pricing and editions	Not disclosed
Usage limits	Not disclosed

........

In Tableau Next, the announcement says apps live inside workspaces, inherit Salesforce data access controls and are built on Tableau Semantics. On Tableau Cloud and Server, apps live inside workbooks and query Published Data Sources or Embedded Datasources.

··········

BUILDING A DATA APP WITH CLAUDE OR CHATGPT THROUGH TABLEAU MCP.

The Get Started page describes a workflow in which the AI assistant writes and packages the app while the analyst describes it and reviews the result. The assistant needs an MCP connection to Tableau, the Tableau plugin and code execution, because it works with local files and runs the plugin's scripts.

On Tableau Cloud, the connection goes to the hosted Tableau MCP service at https://mcp.tableau.com, added in the AI client's MCP or connector settings and completed with a Tableau sign-in. The hosted service uses OAuth 2.1, and every user signs in with their own Tableau Cloud identity, so tool calls run with that user's permissions. The Tableau plugin, from the tableau/tableau-plugin repository, provides the data app authoring skill. For Claude desktop the documentation gives a build script that produces a ZIP to upload under Customize, Plugins, Add, Upload plugin; for Codex or the ChatGPT desktop app a separate script installs the plugin, after which Tableau is connected from the Plugins screen.

The MCP tools involved are scaffold-data-app, request-workbook-upload and publish-workbook, together with tools that find data sources and read their metadata. The documented sequence runs from checking access and tools, to scaffolding a starter workspace, connecting a published data source, writing the interface with runtime queries and filters, and packaging the result as a .twbx that is published through Tableau MCP to the chosen project. The documentation recommends describing purpose, audience, data source, metrics, controls, visual style and destination. An illustrative request:

Use the Tableau plugin's data app authoring skill.
Build a "Regional Sales Explorer" for sales managers from the published
data source "Sales". Inspect the data source metadata before choosing fields.
Show revenue, order count and average order value by region and month,
with a region filter and a date range control. Query data at runtime.
Include loading, empty and error states. Publish to the project "Sales Apps".

Review stays with the builder. Tableau's guidance is to open the app in Tableau Cloud, test it with real data and every filter, check loading, empty and error states, test access as an intended viewer, and review the app's code and dependencies before sharing. Changes are requested in the same conversation and republished over the existing workbook.

··········

PERMISSIONS NEEDED TO CREATE AND VIEW TABLEAU DATA APPS.

The overview page lists separate requirements for builders and viewers. A data app runs as the viewing user, and the documentation repeats that being able to open an app does not grant additional permissions to its data sources.

........

Role	Requirement listed by Tableau
Builder site role	Creator or Explorer (can publish)
Builder project access	Publish (Overwrite to republish)
Builder data access	View, Connect, API access per source
Builder admin dependency	REST paths and methods allowed
Viewer workbook access	View, full data query, API access
Viewer data access	View, Connect, API access per source

........

An app that works for its author can fail for a viewer who lacks API access to one of the underlying data sources, so test accounts that mirror each audience belong in the review step. Row-level security rules should stay in the data layer, where Tableau applies them to every query the app sends.

··········

ADMIN SETTINGS AND SECURITY CONTROLS FOR DATA APPS ON TABLEAU CLOUD.

Site administrators control data apps from Settings in Tableau Cloud, on the Packages tab. The site-wide switch is Allow Packages on this Site. In the Packages view, each package can be disabled individually from the Actions menu, and the table shows a Version column for each package.

Two further controls decide what an app can reach. Extension Package Allowed Origins lists the external origins that packages may load from, and these must use HTTPS; a package that declares a source outside the list will not run. Under Tableau REST API Access, clearing Allow all REST API access (all paths and methods) blocks any request that is not on the administrator's list of allowed paths. A separate setting governs whether packaged extensions can trigger popups. On the hosted MCP side, administrators cannot switch the service off, but they can exclude tool groups for their site with the EXCLUDE_TOOLS setting through the REST API.

The security page sets the limits of the model. App code executes in the viewer's browser, so front-end resources must not contain passwords, access tokens or other secrets. Declared REST API access does not give the viewer additional Tableau permissions, and external services an app calls are separate dependencies that Tableau does not host. Tableau recommends approving only the origins an app needs, bundling resources where practical, testing with restricted accounts to confirm that failed queries do not expose sensitive details, and repeating the review whenever data sources, dependencies or API usage change.

··········

HOW DATA APPS COMPARE WITH OTHER AI ANALYTICS AGENTS.

Other data platforms are moving assistants onto governed data in a similar way. The OpenAI Data agent in ChatGPT Work queries warehouses such as BigQuery, Snowflake and Databricks and produces interactive dashboards inside ChatGPT, while Snowflake CoCo for Google Sheets runs live queries under the user's Snowflake role. Tableau's version places the generated application inside Tableau itself, so the published app is governed content in a project, with the same permissions, data sources and admin controls as a workbook.

For teams already using MCP with coding assistants, the setup will look familiar: the agent gets tools from a server and works on local files, as described in our overview of Claude Code MCP integrations for databases and external tools. The difference is the destination, since the output is a .twbx published to Tableau Cloud and reviewed by site administrators.

··········

TABLEAU DATA APPS BETA: WHAT TO CHECK BEFORE A PILOT.

Tableau confirms that Data Apps are in beta, that they can be built in Tableau Studio or by MCP-connected agents such as Claude, ChatGPT and Codex, that apps run as the viewing user with object-level and row-level security applied, and that database credentials stay inside Tableau. The developer documentation confirms the package format, the queryAsync(), readMetadata() and fetch() methods, the site role and permission requirements, and the Packages, Allowed Origins and REST API Access settings.

The platform depends on which source a team follows. The beta documentation limits the feature to Tableau Cloud, while the announcement lists Tableau Server and Tableau Next as publishing destinations; Tableau Server customers also need a self-hosted MCP server. Which tools an agent can use depends on the site's SKU entitlements, the user's permissions and any tool groups the administrator has excluded.

Tableau has not disclosed pricing, required editions, package size limits, row or payload limits, or a date for general availability. A sensible pilot starts with one published data source with row-level security already in place, an administrator who has enabled packages and listed the allowed REST paths, and a review of the generated code before the app is shared beyond its builder.

··········

·····

DATA STUDIOS

·····

datastudios.org