Live demo

Try XLTable on a live server — nothing to install

A public XLTable server with a sample retail dataset: ten years of sales for a fictional chain of stores, 3.2 million rows. Connect an Excel PivotTable to it, open the same cubes as a pivot table in your browser, or ask an AI assistant — all three read one semantic layer and return the same numbers.

Connection details · shared login

Server
https://olap-demo.xltable.com/
User name
demo
Password
demo
Database
demo

Everyone shares this login, so treat the server as a public place: reports you save in the browser are visible to other visitors and can be changed by them. Saved reports are removed after 3 hours; the three sample reports below are restored every 3 hours. The data itself is read-only and synthetic — nothing you do can break it.

Three interfaces · same cubes

Pick the tool you already use

Excel · Windows

Excel PivotTable

  1. In Excel: Data → Get Data → From Database → From Analysis Services.
  2. Server name: https://olap-demo.xltable.com/
  3. Choose Use the following User Name and Password and enter demo / demo.
  4. Select the database demo and the cube Demo Sales (or Demo Store Operations), then Finish.
  5. Drag fields into the PivotTable — every refresh is a live query to the server.
Connecting Excel, step by step →

Browser · any OS

Pivot table in the browser

Open the web pivot and sign in with the same login. Start from a sample report or build your own: the field list, Rows / Columns / Values areas, hierarchies and subtotals work like Excel.

Open the web pivot

AI · MCP

Claude and other AI assistants

  1. In Claude: Settings → Connectors → Add custom connector.
  2. Server URL: https://olap-demo.xltable.com/mcp
  3. When the browser asks you to sign in, use demo / demo.
  4. Ask away: “Which region grew fastest last year?”, “Which brand has the highest return rate?”, “Where did sales fall below plan in 2025?”

Any MCP client works the same way — the server speaks the standard protocol with OAuth sign-in.

AI assistants and MCP in the documentation →

What's inside

Two cubes over a fictional retail chain

Demo Sales — sales

Ten years of sales (2017–2026) with cost and profit, customer returns, a monthly sales plan and current stock levels. Dimensions: Geography (Region → City → Store), Products (Category → Brand → Model), Dates (Year → Quarter → Month → Day), plus Manager, Store Format and Price Segment. Calculated fields include Margin %, Return Rate %, Plan Completion % and YoY Growth %.

Demo Store Operations — the stores themselves

Daily store traffic, purchase receipts, staff hours, rent and selling area over the same ten years, with Conversion % and Visitors per Staff Hour. The same Geography, Store Format and Dates as in the sales cube, so results from the two cubes line up by store and period.

Things to look for

The data is generated, but it is not flat — a few stories are planted in it for you to find:

  • ▸ a steady year-over-year growth trend with one bad year in the middle
  • ▸ Black Friday peaks at the end of every November
  • ▸ stores that closed and stores that opened along the way
  • ▸ a brand that appears mid-history and grows from zero
  • ▸ one product with an unusually high return rate
  • ▸ a single-day sales spike in one store and a few days of missing data in one region

Ask the AI assistant to find them, or drill down in Excel — that is exactly what the semantic layer is for.

Run the same thing on your laptop

XLTable Desktop is free — the same semantic layer, Excel, browser pivot and MCP connector on your own machine, pointed at your own warehouse. Ready-to-run sample datasets for every supported database are in the documentation.

Download XLTable Desktop — free

See it on your own data

We can set up a private demo on your warehouse — ClickHouse, Snowflake, BigQuery, Databricks, Trino, StarRocks, Greenplum or DuckDB — with your tables and your access rules.

Request a private demo

The demo server runs on modest hardware and is shared by everyone who visits this page — it is here to show how XLTable works, not how fast it goes. Response times depend on who else is in.