dante-lib

A data science workbench for Claude Code. Gives Claude MCP tools to query databases, build charts, manage knowledge, and create data apps — all from your terminal.

What it does

• SQL tools — execute queries, list tables, describe schemas, profile data
• Knowledge — save validated SQL patterns with embeddings, keyword triggers, and project notes
• Charts & apps — generate Plotly charts and self-contained HTML data apps
• Embedding ingestion — bulk import SQL patterns from Looker, Databricks, or your warehouse
• Management UI — web UI to configure connections, credentials, and knowledge

Install

pip install dante-lib
# or with uv
uv tool install dante-lib

For specific database drivers:

pip install "dante-lib[snowflake]"
pip install "dante-lib[databricks]"
pip install "dante-lib[bigquery]"
pip install "dante-lib[postgres]"
pip install "dante-lib[mysql]"
pip install "dante-lib[all]"   # everything

Requires Python 3.11+ and an OpenAI API key for embeddings.

Quick start

# Create a new project
dante launch my-project
cd my-project

# Or initialize in an existing directory
dante launch

# Open the management UI to configure your database connection
dante ui

Then open Claude Code in the project directory. The .mcp.json file is already configured — Claude will have access to all dante tools automatically.

Project structure

After dante launch:

my-project/
├── .mcp.json                  # Claude Code MCP config (auto-configured)
├── CLAUDE.md                  # Tool reference loaded by Claude each session
├── .dante/
│   ├── config.yaml            # Default connection
│   ├── knowledge/
│   │   ├── keywords.yaml      # Keyword → SQL hint mappings
│   │   ├── notes.md           # Free-form notes Claude reads each session
│   │   └── patterns/          # Saved SQL patterns (.sql files)
│   └── embeddings.db          # SQLite embedding index (gitignored)
├── .claude/skills/            # Slash commands (/query, /dashboard, /analyze, ...)
├── analysis/                  # Analysis scripts
├── outputs/                   # Generated charts, dashboards, reports
└── data/                      # Local data files

MCP tools

Claude gets these tools when it opens the project:

Slash commands

Skills scaffolded into .claude/skills/:

Python library

Use import dante in analysis scripts for DataFrame-based work:

import dante

df = dante.sql("SELECT * FROM orders LIMIT 100")
dante.tables()
dante.describe("orders")
dante.profile("orders")
dante.chart(df, x="date", y="revenue", kind="bar", title="Revenue")

dante.knowledge.search("monthly revenue by region")
dante.knowledge.save_pattern("monthly_revenue", sql="...", description="...")

dante.checkpoint("before-pivot")
dante.rollback("before-pivot")

Embedding ingestion

Pull SQL patterns from BI platforms into the local embedding index:

# From the CLI
dante ingest --source looker
dante ingest --source databricks
dante ingest --source all --dry-run

# Or from the management UI
dante ui

Ingestion pulls dashboard titles and SQL, generates natural-language questions with GPT-4o-mini, creates OpenAI embeddings, and upserts into embeddings.db. Claude then finds these patterns when you call /query.

Connections for Looker and Databricks are configured in the management UI and stored in ~/.dante/credentials.yaml (global, not per-project).

Management UI

dante ui           # opens at http://localhost:4040
dante ui --port 8080

Configure database connections, API credentials, and keywords. Connections are stored globally (~/.dante/) so the same connection works across all your projects.

CLI reference

dante launch [name]          # scaffold a new project (or init in place)
dante ui [--port PORT]       # open the management UI
dante mcp serve              # start the MCP server (called by Claude via .mcp.json)
dante ingest [--source ...]  # run embedding ingestion
dante status [--json]        # show connection, knowledge stats, and output count
dante open [name]            # open a generated artifact in the browser

Configuration

Per-project (.dante/config.yaml):

default_connection: my-warehouse

Global (~/.dante/connections.yaml):

my-warehouse:
  dialect: snowflake
  account: xy12345
  database: ANALYTICS
  warehouse: COMPUTE_WH
  role: ANALYST

Global credentials (~/.dante/credentials.yaml):

openai:
  api_key: sk-...
looker:
  base_url: https://company.looker.com
  client_id: ...
  client_secret: ...

License

MIT