A few days ago, Matthew Miller told me he wanted to see what happens when you combine an open-source Tableau workbook library with an MCP server. Today I can show him.
The demo in this video is a guided agentic authoring run. It starts with a real Excel file. No baked demo data. No hardcoded charts.
The agent calls my MCP server step by step: 1. Reads and summarizes the schema 2. Proposes 2-4 analysis directions → human picks one 3. Drafts a full dashboard contract → human reviews 4. Generates an ASCII wireframe → human approves 5. Builds an execution plan → generates the .twb
The entire run is logged, resumable, and validated against the official Tableau XSD schema — a pointer Matthew himself shared with me.
This is what "Agentic BI" looks like in practice: not AI replacing the analyst, but AI doing the mechanical work while the human holds the steering wheel at every decision gate.
What excites me most about what's next: imagine the agent generating a dashboard while you watch it render live in Tableau Desktop — close the feedback loop in the same conversation, tweak a chart type, adjust a filter, and re-generate in seconds. Add cwprep upstream for automated data prep, and a Tableau Server publish step at the end, and suddenly the entire analytics lifecycle — from raw data to a shared, production-ready dashboard — becomes a single controlled, auditable agentic run.
Matthew, I told you I'd keep you updated on the PoC. Here it is.
Southard Jones Elif Tutuk Matthew Miller Adam Mico Allan Folting Roy Raviv Would love your thoughts!
#Tableau #DataVisualization #ArtificialIntelligence #BusinessIntelligence #OpenSource #DataAnalytics #MCP #AIAgents #DataEngineering #datafam #gemini #google #ai #bi #agent #cwtwb #codex #chatgpt #salesforce