MCP server + CLI

Dolex on the
command line.

Dolex ships as both an MCP server and a CLI — let Claude drive it, or run it yourself with a single dolex command.

The same analysis engine and 43 chart types your AI assistant uses, available straight from the shell. Point dolex at a CSV and get a chart, a SQL answer, or a full analysis plan — built for scripts, CI, and pipelines.

npm install -g @outsidedata/dolexFull CLI reference →

~/data — dolex

$ dolex visualize sales.csv -i "compare revenue by region" ✓ bar "Revenue by region" • Bar Chart (score: 75): default comparison — categories with one numeric value data 6 rows × 2 cols chart ~/.dolex/charts/chart-spec-d6f07a5c.html hash spec-d6f07a5c $ dolex query sales.csv "SELECT region, MEDIAN(revenue) m FROM sales GROUP BY 1" region m────── ─────East 21000West 14300$

Charts default to ~/.dolex/charts/ — pipe-friendly via --stdout / --json.

Two frontends, one core

Two doors to the same engine — your AI assistant drives it in conversation, or you run it yourself from the shell.

For AI assistants

MCP server

Register Dolex once and Claude gains 22 tools — connect CSV files or a live Postgres/Mongo database, query, auto-analyze, and visualize, all in conversation. Runs locally; CSVs never leave your machine, and database queries hit only your own infrastructure.

$ dolex mcp

For scripts & direct use

Command line

Run the same toolkit yourself with a dolex command — composable in shell scripts, CI, and Makefiles, or a one-liner to chart a CSV. Reads a CSV file, a JSON array on stdin, or a registered source.

$ dolex visualize sales.csv -i "trend over time"

Commands

One binary, twelve verbs

◧

Chart

Turn data into a chart for its shape

visualize

Render a chart from your data

Matches a pattern to your data and intent, writes a self-contained HTML chart (optionally a PNG).

e.g.dolex visualize sales.csv -i "revenue by region"

refine

Tweak a chart by its hash

Sort, filter, limit, recolor, switch chart type — across separate invocations.

e.g.dolex refine spec-1a2b --sort desc --limit 10

◔

Explore

Query and understand your data

query

Run SQL, print rows

Full SQL with custom aggregates (MEDIAN, P25/P75, STDDEV). Output as table, json, csv, or ndjson.

e.g.dolex query diamonds.csv "SELECT cut, MEDIAN(price) …"

analyze

Auto-generate an analysis plan

Classifies columns and proposes prioritized analyses with ready-to-run SQL for each.

e.g.dolex analyze diamonds.csv

describe

Profile columns

Types, roles, distributions, stats, and sample rows for every column.

e.g.dolex describe diamonds.csv

check

Audit data quality & footguns

Type traps, sentinels, leaked/identical columns, duplicate rows, outliers. Exits non-zero on HIGH findings.

e.g.dolex check diamonds.csv

◩

Derived columns

Persisted transforms that survive across commands

transform

Add a derived column

Define a new column with an expression; it persists to a sidecar manifest and is visible to later commands.

e.g.dolex transform diamonds.csv --create ppc --expr "price / carat"

columns

List columns by layer

See source / derived / working columns for a dataset.

e.g.dolex columns diamonds.csv

drop

Remove derived columns

Drop derived or working columns in dependency order.

e.g.dolex drop diamonds.csv ppc

◎

Reference & server

Browse patterns, manage sources, run the MCP server

patterns

Browse all 43 chart types

List every pattern by category, or show one in detail.

e.g.dolex patterns

sources

Manage data sources

e.g.dolex sources add sales ~/data/sales.csv

deps

Check environment capabilities

Report which source types and optional features are available here, and how to enable the rest.

e.g.dolex deps

mcp

Run the MCP stdio server

The same binary becomes the MCP server for Claude Desktop / Code.

e.g.dolex mcp

43 chart types

Bump charts, beeswarms, Sankeys, ridgelines, choropleths — Dolex matches the visualization to the shape of your data and the question being asked.

Real SQL engine

In-memory SQLite over your CSVs with custom aggregates, window functions, time bucketing, and joins. Handles 500k+ rows — no code generation.

100% offline

Charts render as self-contained HTML with embedded D3 — zero CDN calls. 33 TopoJSON maps ship with the package. Your data stays on your machine.

PNGHTML charts, querying, and analysis need nothing extra. Rendering a chart straight to a .png (the --png flag) is opt-in — enable it once with npm install playwright && npx playwright install chromium.

Install it and chart something

One global install gets you the dolex command and the MCP server.

npm install -g @outsidedata/dolex