Get started with Dolex

Install Dolex globally with npm:

npm install -g @outsidedata/dolex

This installs the dolex command globally. Next, you'll point Claude at it.

Dolex runs as an MCP server that Claude Code connects to over stdio. You register it once and it's available in every conversation.

Add to your MCP config

Open your Claude Code MCP settings and add Dolex as a server. The exact config depends on your setup:

{
  "mcpServers": {
    "dolex": {
      "command": "dolex"
    }
  }
}

Verify the connection

Once configured, you can verify Dolex is running by asking Claude:

The fastest way to see Dolex in action is to point it at a CSV file on your machine. Tell Claude to load it, then ask a question about the data.

Notice that Claude didn't default to a bar chart. Dolex's pattern selector analyzed the data shape and your comparison intent, then recommended a chart type that actually reveals the story — in this case, the crossing trajectories that a bar chart would hide.

Force a specific chart type

If you want a specific chart type, just ask for it by name. Dolex has 43 chart types — far beyond bar, line, and pie:

Browse all 43 chart types →

This is where Dolex really shines. Instead of asking one question at a time, you can hand it a dataset and let Dolex figure out what's interesting. It examines every column, understands what each one represents, and generates a full analysis plan.

How to use it

Just ask Claude to explore your data. It will load the CSV, call the analysis planner, and execute each recommended visualization with a narrative finding. The output looks like a data analyst sat with you and walked through the dataset.

Load ~/data/sales-2024.csv and explore what's interesting

What the planner does

The planner examines column profiles and generates 4–6 analyses, each with a ready-to-execute query. It picks from these categories:

The planner diversifies — it won't produce two trend charts or two comparisons. Each step covers a different analytical angle.

Manual approach

If you prefer to drive the analysis yourself, you can load a CSV and call the planner separately:

For anything beyond small inline datasets, load a CSV file (or a whole directory of them). Dolex processes the data locally — only summaries and chart specs flow through the conversation, not your raw data. Load multiple CSVs and Dolex can automatically combine them by matching columns.

Load a CSV

Tell Claude about your CSV file. Dolex loads it locally so you can ask questions about it without sending your actual data to the AI:

Once loaded, you can ask follow-up questions without reloading. Loaded CSVs persist across conversations:

What stays private

Every chart Dolex produces can be refined. Ask for changes naturally — Claude will call the refine tool with no data round-trip (it references the chart by its spec ID).

Things you can change

Switch chart types

Every visualization comes with alternative chart type recommendations. You can switch between them without re-sending data:

Color palettes

Dolex ships with purpose-built palettes. Ask by name or by intent: