Labsco
MCP SERVER

MCP Data Visualization Server

by xoniks ↗

Ask for a chart in words and get an interactive one back — over local DuckDB or a Databricks SQL warehouse.

Spreadsheet & Tabular File Analysis
Summary
From a question to a chart without writing the SQL or the plot.

Both halves are handled — the query against your data and the rendering — and because the chart choice comes from rules rather than another model call, the same question gives the same chart and nothing leaves the machine on the DuckDB path.

What it is

A visualization server that pairs a database connection with chart generation. You describe what you want to see, it queries and renders a Plotly chart as an interactive HTML widget, and the chart suggestion is rule-based rather than delegated to another model — so it works offline once configured.

What you get
  • Eight chart types with the case each one serves: bar for comparing categories, line for trends over time, scatter for relationships, pie for proportions, histogram for distributions, box for comparing distributions, heatmap for correlations, area for cumulative trends
  • Data access in plain language — list the tables, analyse one, or query the top rows by a measure
  • DuckDB locally, with CSV import and an interactive database browser, plus the ability to switch to another `.duckdb` file mid-conversation
  • Databricks as an alternative — browse catalogs, list schemas, switch catalog, and check which database you are connected to
  • Automatic statistical analysis and pattern detection alongside the chart
  • A CLI: `mcp-viz configure` sets up the client, `mcp-viz status` reports what is configured, `mcp-viz test` exercises the server, `mcp-viz create-db` builds a sample database, and the `mcp-viz databricks` subcommands configure, test and remove credentials
Requirements

`pip install mcp-visualization-duckdb` brings DuckDB and everything else with it — no separate database install. Then `mcp-viz configure` detects your platform, finds the client config, backs it up, writes the entry and validates the result. For Databricks you need the workspace hostname, the SQL warehouse HTTP path and a personal access token, entered through `mcp-viz databricks configure`, which stores them in your system keyring or an encrypted file rather than in plain text or shell history. Setup is one-time; you reconfigure when a token expires or you move workspace or warehouse. SQL injection protection is built in.

Setup effort

One command — pip install mcp-visualization-duckdb && mcp-viz configure