ade-catalog — ADE (Agentic Data Engineering)

A local, self-hostable metadata catalog for Databricks + Power BI — with cross-platform lineage, an interactive graph explorer, and an MCP server so Claude can query your data estate.

Preview edition. This is the open, self-host preview: SQLite, local, no login, Power BI + Databricks from files you already have.

What it does

• Parse Power BI (TMDL / PBIP projects) and Databricks (.py notebooks) from files on disk
• Catalog every table, measure, notebook, report across platforms — browse + full-text search
• Trace cross-platform lineage in an interactive graph (bronze → gold → semantic model → measure)
• Ask — an MCP server exposes the catalog to Claude Desktop / Claude Code

Quick start

git clone https://github.com/rbutinar/ade-catalog.git
cd ade-catalog
pip install -r requirements.txt          # SQLite = stdlib; no DB drivers to install

# 1. Create the local SQLite catalog
python -m ade_catalog.schemas.deploy_sqlite

# 2. Build the demo catalog (synthetic Acme estate)
python -m ade_catalog.ingest --source demo
python -m ade_catalog.cli lineage build --view Demo

# 3. Browse it
python -m ade_catalog.cli serve            # http://localhost:5001

Everything runs locally against a SQLite file — no accounts, no cloud, no login.

Use it with Claude (MCP)

The bundled .mcp.json starts the catalog's MCP server over stdio — just open the project in Claude Code, or add it to Claude Desktop:

{
  "mcpServers": {
    "ade-catalog": {
      "command": "python",
      "args": ["-m", "ade_catalog.mcp_server.stdio"],
      "cwd": "/path/to/ade-catalog"
    }
  }
}

Then ask: "What measures depend on the sales fact table?", "Trace the lineage from bronze to the Power BI model", "Which notebooks write to gold?"

Bring your own data

Drop your exported files under ade_data/<env>/inputs/ and rebuild:

• Power BI — a PBIP project's definition/ folder (TMDL): tables, columns, measures (DAX), relationships
• Databricks — exported .py notebooks: table lineage (inputs/outputs) with variable resolution

python -m ade_catalog.ingest --source ade_data/my_env/inputs --view "My Env"
python -m ade_catalog.cli lineage build --view "My Env"

This repository vs. the hosted app

	This repository (public core)	Hosted app (sign-in)
Backend	SQLite, local	Postgres / SQL Server, multi-tenant
Sources	Power BI + Databricks, from files	+ Fabric, Qlik, Tableau, SAP BusinessObjects and BW, Oracle, SSIS, Informatica, Talend, Synapse, PostgreSQL, dbt, Cloudera, … + live extraction
Lineage graph	✅ interactive explorer	✅ + large-graph performance
MCP	✅ (stdio + HTTP)	✅ hosted, org-scoped
AI assistant, AI enrichment, Excel/query export, data products	—	✅
Publish & share (multi-user, RLS)	—	✅

Not everything is public. Some parts stay private for reasons of security and confidentiality, some were developed under specific commercial agreements, and some are not released yet. The public core is what you build on: the data model, the connector pattern and the MCP interface. Add a connector, adapt it to your estate, contribute back.

A hosted demo runs at ade-catalog.vercel.app: sign up; with an invitation code you get the demo at once, otherwise you join the waiting list. More about the family: ade-family.com. Contact: [email protected].

License

Apache 2.0 — see LICENSE.

Roberto Butinar — LinkedIn · GitHub