Skip to main content

Less than three weeks before Data Outpost. Get your ticket before they sell out.Join AI and Data builders Nov 4-5 in SF

GO BACK TO BLOG

MotherDuck is coming to Azure

SqlIntegrationsProduct

· 5 min read

BY
Jacob Matson

TL;DR: MotherDuck is coming to Microsoft Azure. Join the waitlist and tell us which Azure region you need. You don't have to wait to start on the AI side: the MotherDuck MCP server already works with Copilot Studio and GitHub Copilot.

Many data teams don't get to pick their cloud. Their company standardized on Microsoft years ago, so identity runs through Entra ID, the data is on Azure Blob Storage, and the reports run through Power BI. Hundreds of those teams have asked us for MotherDuck on Azure. Today they query across clouds from MotherDuck on AWS, and Azure bills them for egress on every query.

So we're bringing MotherDuck to Azure. We're building it to be the same product, running in Azure instead of AWS.

What MotherDuck is

MotherDuck is a serverless data warehouse built on DuckDB, where every person and agent gets its own compute on the same data.

Most warehouses give each team or workload its own cluster, sized by an admin. That works when a few analysts run reports. It breaks down when every person, and every person's agents, ask questions at the same time. Sizing each cluster for the peak leaves capacity that is mostly idle, and sizing it for the average means people wait at the peak.

Instead, MotherDuck gives every user, every agent and every customer its own compute, called a Duckling. Each one connects to the same shared data and is sized independently. A heavy agent doesn't slow anyone else's work. These Ducklings scale to zero when they're idle, so you pay for what runs, not for the peak.

On top of that architecture, we've built the rest of the platform for agents.

Built for agents

  • An MCP server for your agents. Connect Claude, ChatGPT, Cursor, Copilot and other MCP clients to your MotherDuck databases, and let the LLM write the SQL query. There are both read-only and read-write MCP tools depending on your needs.
  • Dives: data apps, built by agents. Ask a question and get back an interactive, live app. Share it with your team, or embed it in your own product for your customers.
  • Guides: business context in one place. Write down how your company defines revenue, which table to trust and what's changed recently, so every person and agent works from the same definitions.
  • Flights: pipelines next to the data. The agent writes Python jobs for ingestion, enrichment and transformation of data, with a built-in cron for scheduling.

The data team uses these to bring in the data and write down what it means. Everyone else asks questions on top, each person and agent on its own Duckling, so a busy agent never slows anyone else down.

What it looks like in practice

AheadComputing, a startup designing a RISC-V CPU, built its first data platform on Azure Postgres. Once the team started using it, transformations on small data sets took over an hour. It moved to MotherDuck and was in production within a week. Hardware engineers who don't write SQL now ask Claude Code, which queries MotherDuck through the MCP server. 74% of the company's engineers were onboarded in the first weeks. A dashboard request that used to take three months now takes about 90 minutes, because engineers build their own Dives.

Azure Hero — Meadow (photo)@2x.png

Ready for Microsoft

The Microsoft tools your company already uses connect to MotherDuck today.

  • Entra ID. Your team signs in to MotherDuck with the Microsoft identity it already uses.
  • Copilot Studio. Add MotherDuck as an MCP tool on your agent. Each person can sign in with their own MotherDuck account, or the agent can use a shared service account. For an agent you publish to a wide audience, a read-scaling token keeps it read-only.
  • GitHub Copilot in VS Code. Install the MotherDuck MCP server, and its tools appear in Copilot's Agent mode.
  • Power BI. Connect through Power BI's native PostgreSQL connector and the MotherDuck Postgres endpoint. You don't need a custom connector or a driver install.

The MCP setup docs and the Power BI guide have the detailed steps, and today your Copilot agent and your Power BI reports reach MotherDuck on AWS. With Azure regions, the compute will sit in Azure, right next to your data.

In addition to handling data warehousing workloads, MotherDuck can sit beside the other parts of the ecosystem. Fabric, Synapse and Databricks can handle portions of the work, with MotherDuck serving the interactive layer: dashboards, ad hoc analysis, customer-facing analytics and agents.

Help decide which region we start with

We haven't picked our first Azure region yet, and we want feedback from customers and prospects to decide. Regions will open in order of demand, and early access within each region goes out in waitlist order. We'll email you when yours opens, and we'll share timing, pricing and what early access includes before then. Joining doesn't commit you to anything.

In the meantime

MotherDuck on AWS can already read Parquet and CSV files in Azure Blob Storage:

Copy code
CREATE SECRET IN MOTHERDUCK (TYPE azure, CONNECTION_STRING '<your connection string>');

SELECT * FROM 'az://my-container/events/*.parquet' LIMIT 10;

The Azure Blob Storage docs have the details.

Join the waitlist and tell us which Azure region you need.

Table of contents

Subscribe to motherduck blog