Find wasted DBUs from your Databricks system tables.

DBU Waste Finder is a free, open-source notebook (and an MCP server) that reads your Databricks system tables and lists which jobs, clusters and SQL warehouses cost the most and which look wasteful, with the DBUs involved and, where system.billing.list_prices has a price, an estimate at list price. Read-only. The notebook keeps everything in your workspace; the MCP server sends results to your AI client (see Privacy).

Get the free notebook

What it checks

Top cost drivers
DBUs and list-price cost by product, job, warehouse and cluster.
Jobs on all-purpose compute
Scheduled jobs running on interactive clusters, with the list-price difference to job compute.
Cluster auto-termination
All-purpose clusters with auto-termination off or above 60 minutes.
Idle or oversized clusters
Clusters with low average CPU and memory use.
SQL warehouse auto-stop
Warehouses with auto-stop off or above 60 minutes, including ones deleted in the period.
Idle SQL warehouses
Warehouse uptime with no query running, in DBUs.
Serverless jobs
DBUs per serverless job and the share spent on failed runs.

Classic-cluster checks are tested on mock data; the warehouse and cost checks have also been run against a live trial workspace. Costs are list price, not your discount. Some SKUs have no list price; then the tool shows DBUs and marks USD as unknown.

Pro

Pro, waitlist (around $149/month per workspace). Not built yet – email mark@dataandbeyond.dk to join the waitlist.

Privacy

The notebook: runs inside your workspace with read-only SELECT on system tables (billing and compute metadata, not your business data). Nothing is sent anywhere.

The MCP server runs on your machine, not in your workspace. It queries your system tables with a token you provide and passes the results to your AI client (e.g. Claude or Cursor), and through it to that AI provider. Results can include job, cluster and warehouse names and ids, DBUs, list-price amounts and owners' email addresses. Use a read-only service principal and only connect it to an AI client you are allowed to share that metadata with.

Nothing is sent to us. If you email us, we only keep your name and email.

Data and Beyond ApS · CVR 42397075 · Denmark. Not affiliated with Databricks, Inc.