MotherDuck + Go Fig

Database

Connect MotherDuck to Go Fig for serverless cloud analytics powered by DuckDB.

MotherDuck is DuckDB in the cloud: fast analytical queries, no cluster to manage, and a local-first model that lets you blend cloud data with files on a laptop. Go Fig connects to MotherDuck as a first-class node in the Financial Intelligence Graph so Celeste and AI financial analysts can query shared databases, attached S3 or GCS-backed tables, and custom SQL views against the rest of your GL, CRM, and operational data. The connector uses MotherDuck's service-token authentication, supports both cloud-hosted and hybrid (cloud + local) databases, and takes advantage of DuckDB's columnar execution for fast extracts. Parquet and CSV files referenced via read_parquet() or read_csv() work transparently, as does MotherDuck's share feature for cross-account data sharing. This is a particularly good fit for lean finance teams that want warehouse-grade analytics without paying for a Snowflake or BigQuery footprint, and for data teams standardizing on DuckDB in local development.

Connect MotherDuck See All Integrations

Key facts

Compute: Serverless DuckDB in the cloud
Auth: Service token (read-only)
File access: Parquet, CSV from S3 / GCS / HTTPS
Sharing: MotherDuck shared databases
Hybrid: Local DuckDB + cloud tables in one query

SOC 2 Type II · All integrations

What you can do with MotherDuck data in Go Fig

DuckDB-first analytics stack

If your data team uses DuckDB/MotherDuck as the lightweight analytical layer, Go Fig plugs into it directly so finance and data share one source of truth.

Parquet lake queries without a warehouse

Use MotherDuck's read_parquet() against S3 or GCS Parquet files as the query layer, and let Celeste join those results to QuickBooks and Salesforce live in Go Fig.

Fast ad-hoc analysis and prototyping

Spin up a MotherDuck database, load a few files, and have Celeste answer finance questions in minutes without standing up Snowflake or Redshift.

Data available from MotherDuck

Go Fig extracts and normalizes the following data from your MotherDuck account:

MotherDuck databases

Tables and views

Shared databases (from other accounts)

read_parquet() external files

read_csv() external files

HTTPS-backed datasets

DuckDB extension outputs (httpfs, json, etc.)

Custom SQL queries

Hybrid local + cloud tables

CTE / macro outputs

How to connect MotherDuck

Generate a service token

In the MotherDuck UI, go to Settings > Service tokens and create a read-only token scoped to the databases Go Fig should access. Give it a descriptive name (e.g., fig-prod-reader) so rotation and revocation are easy. Copy the token; it's shown once.

Point Go Fig at your MotherDuck database

In Go Fig, paste the token and specify the database name. Go Fig enumerates schemas, tables, views, and attached external databases (e.g., databases shared with you via MotherDuck's share feature). No VPC or network setup is required since MotherDuck is a managed service.

Register any file-backed tables

If your analytical patterns use read_parquet() or read_csv() from S3, GCS, or HTTPS, those queries work through Go Fig as-is. For frequently-queried external files, create a view in MotherDuck (CREATE VIEW ... AS SELECT * FROM read_parquet(...)) so Go Fig sees a stable schema instead of re-introspecting files every sync.

Pick sync strategy per table

MotherDuck tables sync on a cadence you set (default 15 minutes). For small tables, full refresh is fast and simple. For large tables, use a high-watermark column (last_updated) for incremental syncs. DuckDB's columnar reads make even moderate-size full refreshes inexpensive, so don't over-engineer incremental logic on smaller tables.

Authentication: MotherDuck service token scoped to specific databases, read-only. Tokens are generated from the MotherDuck UI under Settings > Service tokens and can be rotated independently per connection. Go Fig stores tokens encrypted at rest and they never appear in logs or query history exposed to users.

Common Questions About MotherDuck Integration

Is MotherDuck production-grade for finance analytics?

For mid-market finance workloads, yes. MotherDuck inherits DuckDB's analytical performance, handles the data volumes most finance teams care about, and gives you SQL semantics any finance SQL user already knows. It's a particularly good fit for teams that don't need the concurrency profile of Snowflake or BigQuery and want much lower warehouse spend.

Can Go Fig query hybrid (local + cloud) MotherDuck setups?

Go Fig connects to the cloud side of MotherDuck. If your analysts work in a hybrid pattern locally (laptop DuckDB + cloud tables), the cloud portion is visible to Go Fig; purely local tables on a laptop are not. For shared reporting, publish to the cloud database so Celeste and everyone else can reach it.

How does Go Fig handle Parquet files referenced via read_parquet()?

Transparently. The query runs inside MotherDuck's compute, reads the Parquet from the referenced S3 or GCS URL, and returns rows to Go Fig. For production reliability, wrap the read_parquet() call in a view so the schema is stable. For direct file-lake patterns without MotherDuck, use Go Fig's Amazon S3 or Google Cloud Storage connector instead.

What about MotherDuck's shared databases?

Shared databases appear as attached catalogs in your MotherDuck instance, and Go Fig enumerates them the same way as your own databases. This is useful for consuming data published by another team or a vendor partner without copying it. Permissions on shared databases are controlled by the share publisher.

What are the practical concurrency limits?

MotherDuck's serverless compute scales per-query rather than per-cluster, so finance-scale workloads don't typically hit contention. If you're running many concurrent dashboards plus heavy syncs, MotherDuck's pricing tier determines the cap. Go Fig queues sync jobs respectfully and uses incremental refreshes to keep pressure on the warehouse low.