LoginSign Up
← Back to all sparks
M

MotherDuck

ANALYTICS
Velocity5.0

Serverless cloud data warehouse and analytics built on DuckDB

motherduck.com ↗

MotherDuck is turning the warehouse into an agent's runtime, not just its query target.

agent-nativemcpicebergpostgres-compatibilitydata-appsenterprise-controls
◆Current state
MotherDuck ships roughly weekly, and the notes divide cleanly into three tracks: an agent-facing surface (Dives, Flights, MCP), open-catalog interop (Iceberg on Databricks, Cloudflare R2, DuckLake), and enterprise plumbing (SCIM, RBAC, four AWS regions). Dives — data apps built from natural language — went GA in June and now carry a draft/ready/endorsed/archived lifecycle exposed through MCP and SQL. Flights, a Python pipeline runtime any MCP agent can create, went from Business-only preview to every plan inside a month.
◆Where it's heading
The Postgres wire endpoint is doing the work a driver team would otherwise do: Looker, Drizzle, dbt Cloud, DBeaver and DataGrip all landed through it rather than through bespoke connectors. On storage, MotherDuck is decoupling from its own catalog — server-side Iceberg attach, Databricks-managed table writes and R2 Data Catalog all point at being the compute layer over data it doesn't hold. The agent track is taking the most surface area, with Dives now rendering inside ChatGPT and Claude Cowork and MCP responses re-encoded in TOON to cut payload size.
◆Prediction
Expect Flights to reach GA and to fill in the Sydney and Tokyo regions it currently skips, and the Iceberg REST attach to graduate out of Preview with write support matching the Databricks and R2 paths.

◆Recent moves

  1. 7d ago

    RBAC lands; Cloudflare R2 Iceberg catalog read/write

    Role-based access control on Business and Enterprise closes the last obvious enterprise gap after SCIM and manual deprovisioning. R2 Data Catalog support extends the pattern of treating other vendors' Iceberg catalogs as first-class storage, and a max-runtime field finally gives Flights a cost ceiling.

    View source ↗
  2. 14d ago

    Sydney and Tokyo regions; Flight scheduling on Lite

    Two more AWS regions extend the data-residency story that started with Oregon and Dublin, though Flights don't run in them yet. Moving scheduled Flights down to Lite without per-user or concurrency limits continues the pipeline runtime's march out of the enterprise tier.

    View source ↗
  3. 21d ago

    Flights open to all plans; Databricks Iceberg writes

    One month after its Business-gated preview, Flights is on Free and Lite — MotherDuck is treating the pipeline runtime as part of the product rather than an upsell. Writing to Databricks-managed Iceberg tables and adding a draft/endorsed lifecycle to Dives both point at governed artifacts shared across a team.

    View source ↗
  4. 28d ago

    Server-side Iceberg catalog attach enters preview

    Attaching an external Iceberg REST catalog server-side lets MotherDuck query tables it doesn't store, alongside its own and DuckLake data. Connecting dbt Cloud through the Postgres endpoint removes a self-hosted pg_duckdb proxy from the path — the same substitution the endpoint has been making for BI tools.

    View source ↗
  5. 1mo ago

    Postgres endpoint fixes schema metadata for BI tools

    Resolving primary keys, indexes, column precision and comments is unglamorous work, but it is what makes DBeaver and DataGrip read a MotherDuck schema correctly — the endpoint only substitutes for a real driver if the metadata is right. REFRESH DATABASE on read-scaling replicas lets embedded Dives pull fresh writes on demand.

    View source ↗
  6. 1mo ago

    Flights: Python data pipelines any MCP agent can run

    ⚡ SPARK

    This is the entry the last six weeks descend from: a Python pipeline runtime inside the warehouse, plus Dives reaching GA. Everything since — Flights on every plan, scheduling, per-run config, runtime caps — is distribution and hardening on top of it.

    View source ↗