// Tableau + AI
A working guide for Tableau and BI developers: building extensions with Claude, connecting Claude to Tableau, and the day-to-day of writing, reviewing and automating Tableau work with AI. Written from building it, not from reading about it.
// Build with Claude
Extensions, MCP and skills: the hands-on side. Prompt it, get working code, load it in Desktop.
The full walkthrough: the prompt, the Extensions API code, the .trex manifest, local testing, loading it in Tableau Desktop and hosting it. Plus the shortcut where Claude writes the component and TableauOps hands back the .trex.
Tableau's own MCP server connects Claude to your content and data; the TableauOps connector lets Claude publish viz extensions. What each is for, setup in Claude and Claude Code, example prompts, and how to keep it read-only.
Which Claude Code skills actually help Tableau work, how to install them, what the free Tableau extension skill contains, and how to write one for your team's own standards.
A copyable prompt library: a base prompt plus eleven chart prompts, each tied to a finished extension in the gallery, and the five things every good Tableau extension prompt spells out.
// Work with AI
Where AI saves real hours in Tableau work, where it quietly gets things wrong, and how to check it.
An opinionated line for BI and Tableau developers: hand AI the calc drafts, refactors, docs, QA lists, extension code and REST scripts. Keep data modeling, metric definitions, governance and the final review. Plus a weekly workflow.
Prompts that give the model your schema and grain, worked FIXED, INCLUDE, EXCLUDE, table calc, date and null examples in valid Tableau syntax, the nine mistakes LLMs make in Tableau calcs, and a checklist to verify them.
What's inside a .twb, what an LLM can and can't judge from the XML, a review prompt that keeps the model honest, a performance and maintainability checklist, and two free tools: the Workbook Roast and the Viz Assay.
Use Claude to write TableauServerClient scripts: PAT sign-in, stale content, failed extract refreshes, a permissions audit and a dry-run bulk ownership change, plus the safety rules I follow before anything touches production.
// Reference
Which AI tool fits which Tableau job, and plain definitions for the words that keep coming up.
Claude, ChatGPT, Tableau Agent, Pulse and Copilot compared for the six jobs a Tableau developer actually does: calcs, extensions, admin scripts, workbook review, end-user Q&A and documentation. Honest verdicts, including where Tableau's built-ins win.
Short, accurate definitions of the terms that come up when you use AI with Tableau, from Tableau Agent, Pulse and VizQL Data Service to MCP, skills, tokens and the .trex manifest, each linked to official docs and a deeper guide.
// Try it
// Build log
Real sessions, real prompts, and what broke along the way.
You have a .trex and a folder of files, and to use them they have to live on the internet. Here's the part nobody explains: why hosting exists, the one hard rule (HTTPS with a real certificate), and the two ways to get it done.
Skip the table calcs and floating containers. Here's how to build a reusable Tableau KPI card as a custom extension with Claude Code, in one prompt, and when the native route is still fine.
A Claude Code skill is the memory you hand the AI before you prompt: your rules, patterns, and fixes, baked in. Here's what a skill actually is, why it beats a better prompt, and what changes when you use one.
I gave Claude Code the same prompt twice: once with a Tableau extensions skill, once without. Same model, same data. One fought the manifest for 40 minutes. The other built a working KPI card on the first prompt. Here's why.
