Tableau MCP enables users of Tableau Server and Tableau Cloud to have AI perform actions against the content stored in their Tableau environment. The most common application is a 'chat with your data' chatbot that would allow users to ask questions like:
"How would we improve performance in the business? What's working well, what areas need improvement, what lessons can we learn from the past?"
An AI model would then be able to craft a response to this using tools to interact with the content available in your Tableau Environment. Tableau MCP provides several tools for an AI to use.
In our ‘chat with your data’ example, the tools would help:
Find a list of the data sources and pick a relevant one for the question
Read the metadata of the data source to see its columns and descriptions to see what will answer the question
Execute queries against the data source to produce the analysis required
This AI application would then be able to handle stakeholders' ad hoc requests via a desktop application such as Claude Desktop, a webpage or contained within a Tableau dashboard.
Flexibility is a key benefit of Tableau MCP.
Bring any AI model you want
Configure the style and tone of your AI
Choose the tools you want for your use case
Stay within your security requirements
Apply logging to understand the use of your app
Seeing Tableau MCP in action is incredibly helpful for understanding how it works. In August, I spoke at the Swisse Romaine TUG, with demos in Claude Desktop, a webpage, and a dashboard extension.
The demos start at 1:03:00
The content from this talk is publicly available, so you can start testing Tableau MCP in your own Tableau environment.
QuickStart guides
These guides get you started running Tableau MCP via a desktop application.
Tableau MCP Starter Kit
The starter kit lets you host Tableau MCP in a web app and a Tableau dashboard extension for distributing a 'chat with your data' app to your organisation. The code is available on GitHub with a video tutorial walking through the setup and how the code works.
Tableau MCP Webinar Wednesday 8 October 2025
If you would like to hear more about Tableau MCP and its roadmap, you can attend our upcoming webinar:
This next section goes further into how to develop, customise and get the most out of the Tableau MCP functionality.
Overview of Tableau MCP Tools
For developing your use case, start by understanding the tools available from Tableau MCP. The tools can currently be grouped into three key areas:
Data Sources: find data sources, retrieve their metadata, and execute queries against them
Workbooks: find workbooks, retrieve data and images from workbook views
Pulse: find pulse metric definitions, see users' Pulse subscriptions and retrieve the insight bundle from a Pulse metric
Full details on the tools are on the Tableau MCP GitHub and their documentation site
More tools will be added over time, so this list is likely to become refined and expanded in the future. Tools from other MCP systems can be included too, so whilst Tableau MCP doesn't currently create charts, another MCP could bring this functionality to your application.
Another use case could be that a Tableau centre of excellence/enablement has released a new style guide for dashboard creators. How well has the style guide been adopted? Using the tools for workbooks, they could:
Use list-workbooks to find all their workbooks
Then, get-workbook retrieves the metadata and filters for recently updated views
And get-view-image to download the dashboard images.
Running an AI to compare the dashboard views to the style guide to produce a report of style guide uptake by dashboard creators. Combining this with the dashboard views from Admin Insights produces a priority list of users to contact.
Picking your AI Model and Tailoring Responses
The AI model plays an important role as the point of contact for the user and the decision maker of what tools should be used to answer the user's request.
Whilst the majority of 'off-the-shelf' LLMs will produce a good chat response to a human, a level of sophistication is required for tool calling. Firstly, legacy models lack the internal logic and output structure needed to run a tool call. Secondly, having a large context window is essential for receiving responses from the tools, having ongoing conversations with the user, and performing multi-step workflows.
Start with a more recent model with a large context window. Here, you can establish the typical token usage and costs for an individual user, and use this as a baseline to compare to cheaper models. In my experience, some cheaper models can end up costing more due to repeated runs of tool calls, either failing and/or increasing the amount of tokens in the context window.
Model performance can also be improved by adjusting the system prompt, enriching Tableau content, and reducing the number of tools available to the AI. To enrich the Tableau content, first check what the tool calls are returning. In the case of list fields, which returns the column metadata from a data source, looking at the code associated with the tool call, you can see that descriptions of columns are returned. Adding context to these columns can help an AI better understand some of the synonyms of fields and any particular nuances.
Lastly, you can verify what works best for your use case by establishing a test to judge which AI and performance tweaks work best. From previous tests of AI models, questions that result in a single result are the quickest check for a model accuracy comparison. However, you should also consider the style of responses in more open-ended questions, as some models will provide lots of additional actions for the user, such as a 5-week implementation plan, which may be unnecessary or unrealistic given your circumstances.
Governing AI Access and User Activity Monitoring
Tableau MCP currently authenticates using a personal access token (PAT), which will provide the tools used by the AI model permissions to access the same content on a Tableau site as the user associated with the PAT. Effectively, if you create a PAT token in your setting, you will give the AI model access to all of the data sources and workbooks you have access to on that site.
There are ways you can reduce what the AI can access in the following methods:
Create a new Tableau Site for just the content you want to be accessible to an AI.
Creating a specific MCP user for the PAT and assigning it permissions to the content it should access.
Customising the Tableau MCP tools to only access specific content, e.g. by identifier or by tag
Another aspect to consider is how Tableau MCP is being used. Tools such as Langfuse and LangSmith provide an interface to see interactions with your AI model. Here you'll be able to see conversations users have with the chatbot, the tool calls being run, and the token usage with costs.
This data can help establish how Tableau MCP is being used, leading to the development of additional MCP tools or Tableau content, such as a new dashboard to answer a frequent question.
Tableau MCP offers a great opportunity to explore, understand and deploy an AI application to help themselves or their stakeholders do more with the data they know and trust in Tableau.
The QuickStart guides offer an easy access point to trying out the tools.
Whilst the Tableau MCP Starter Kit shows you how to deploy the chatbot to a web application, into Tableau via a dashboard extension, and enables the customisation of the application for end users.
If you or your team would like a demo of Tableau MCP to understand how it could work in your organisation, please reach out to your account manager or contact us via the website: https://www.theinformationlab.co.uk/contact/