If you've ever opened Tableau Desktop with a clean CSV and then lost an hour to clicking — dragging pills onto shelves, fixing an axis, rebuilding the same profit-ratio calc for the hundredth time — this post is for you.
TabGen is a small, free Windows app that closes the gap between "the data is ready" and "the dashboard is ready." Drop in your data, describe what you want in plain English, click Generate, and you get a real, editable Tableau .twbx workbook. You can grab it at tableaugen.com.
TabGen turns your data — a CSV, an Excel file, or a database connection — into a packaged Tableau workbook using AI. The important details:
localhost:8000. Nothing to configure..twbx.
You describe the dashboard the way you'd describe it to a colleague:
Executive sales overview: KPIs for total sales and profit,
sales trend over time, top 10 products, and sales by region.
…and TabGen builds the KPI cards, the time-series line, the top-N bar chart, and the regional map — arranged on one dashboard. No special syntax. See more examples on the site.
True — and I built one of those too (it's called Twilize). The Model Context Protocol is great. So why build an app on top of it? Because an MCP server and a tool like TabGen solve genuinely different problems, and it's worth being clear about which one you actually need.
An MCP server exposes a set of tools to a language model — list_datasources, run_query, get_workbook — and the model decides, turn by turn, which to call. It's a conversation. That's fantastic for exploration and Q&A over data that's already in Tableau: "which region is dragging down Q3 margin?" is a perfect MCP question.
But building a polished dashboard isn't a Q&A task — it's a construction task with dozens of interdependent decisions. That's where the bare-MCP approach starts to strain:
MCP operates on a running Tableau Server/Cloud site with credentials and published sources. TabGen starts from a file or a database on your own machine and hands you a .twbx. No site, no admin rights.
Ask an MCP agent to "build a sales dashboard" twice and you may get two different tool-call sequences — or a half-built result when the model loses the thread. TabGen wraps the model in a deterministic pipeline: profile the data → parse intent → validate against the real schema → assemble from a proven template → package the file. The model makes the design choices; the scaffolding guarantees a complete, openable workbook every time.
Models hallucinate field names, invent calcs that don't compile, and cheerfully build a map for data with no geography. TabGen checks every step against your actual schema and silently drops anything that fails — so nothing malformed reaches the file.
Editing JSON config, managing a client, handling tokens, prompting an agent through a multi-step build — fine for developers, a wall for the analyst who just has a spreadsheet and a deadline. TabGen is a double-click installer and a text box.
An MCP server gives a model access. TabGen gives you an artifact. One is a protocol for AI-to-tool conversation; the other is a product with a job to finish.
I wrote this up in more depth here: Why TabGen beats a raw Tableau MCP server.
I'm not burying MCP — it's the right layer for conversational analytics over governed Tableau environments, and for agentic workflows you're building yourself. But none of those is "I have a CSV and I want a good dashboard in two minutes." That's the gap TabGen fills.
TabGen is free to download, runs locally, and works with your own Claude or ChatGPT key. Point it at your data, type a sentence, and see what the first draft looks like — you'll know in a minute whether it fits.
👉 Download TabGen at tableaugen.com
If you want to go deeper first:
What would you point it at first? Let me know in the comments.