motherduck

What this skill does

# MotherDuck — Serverless DuckDB in the Cloud


## Overview


MotherDuck, the serverless analytics platform built on DuckDB that combines local and cloud query execution. Helps developers run SQL analytics on cloud-hosted data, share datasets, and build hybrid local-cloud data pipelines using DuckDB's familiar interface.


## Instructions

### Connection and Setup

```python
# Connect to MotherDuck from Python
import duckdb

# Authenticate (opens browser for first-time auth)
conn = duckdb.connect("md:")

# Or with explicit token
conn = duckdb.connect("md:?motherduck_token=your_token_here")

# Create a cloud database
conn.sql("CREATE DATABASE IF NOT EXISTS analytics")
conn.sql("USE analytics")

# Cloud databases persist — accessible from any machine
```

### Hybrid Queries (Local + Cloud)

```python
# Query that combines local files with cloud tables
import duckdb

conn = duckdb.connect("md:analytics")

# Upload a local Parquet file to MotherDuck
conn.sql("""
  CREATE OR REPLACE TABLE events AS
  SELECT * FROM read_parquet('local_data/events_2026_q1.parquet')
""")

# Query cloud data alongside local files
result = conn.sql("""
  SELECT
    e.event_type,
    COUNT(*) AS event_count,
    AVG(e.duration_ms) AS avg_duration,
    u.plan AS user_plan
  FROM analytics.events e                              -- Cloud table
  JOIN read_parquet('local_data/users.parquet') u      -- Local file
    ON e.user_id = u.id
  WHERE e.created_at >= '2026-01-01'
  GROUP BY 1, 4
  ORDER BY 2 DESC
""").df()

print(result)
```

### Data Loading

```python
# Load data from various sources into MotherDuck
conn = duckdb.connect("md:analytics")

# From CSV files (local or URL)
conn.sql("""
  CREATE OR REPLACE TABLE sales AS
  SELECT * FROM read_csv('https://data.example.com/sales_2026.csv',
    header=true, auto_detect=true)
""")

# From Parquet files on S3
conn.sql("""
  CREATE OR REPLACE TABLE events AS
  SELECT * FROM read_parquet('s3://my-bucket/events/*.parquet',
    hive_partitioning=true)
""")

# From JSON files
conn.sql("""
  CREATE OR REPLACE TABLE logs AS
  SELECT * FROM read_json_auto('logs/*.json')
""")

# Incremental loading — append new data
conn.sql("""
  INSERT INTO events
  SELECT * FROM read_parquet('s3://my-bucket/events/2026-03/*.parquet')
  WHERE event_id NOT IN (SELECT event_id FROM events)
""")
```

### Sharing Databases

```python
# Share a database with team members
conn = duckdb.connect("md:")

# Create a shared database
conn.sql("CREATE DATABASE team_analytics")
conn.sql("USE team_analytics")

# Load data
conn.sql("""
  CREATE TABLE monthly_metrics AS
  SELECT
    date_trunc('month', created_at) AS month,
    COUNT(DISTINCT user_id) AS mau,
    SUM(revenue) AS total_revenue,
    COUNT(*) AS total_events
  FROM analytics.events
  GROUP BY 1
  ORDER BY 1
""")

# Share with specific users (via MotherDuck UI or SQL)
conn.sql("GRANT SELECT ON DATABASE team_analytics TO '[email protected]'")

# Recipients connect and query immediately:
# conn = duckdb.connect("md:team_analytics")
# conn.sql("SELECT * FROM monthly_metrics").df()
```

### Node.js / TypeScript Integration

```typescript
// src/analytics/motherduck.ts — MotherDuck in a Node.js application
import duckdb from "duckdb-async";

const db = await duckdb.Database.create("md:analytics", {
  motherduck_token: process.env.MOTHERDUCK_TOKEN!,
});

// Run analytics queries from your API
async function getMonthlyRevenue(months: number = 12) {
  const result = await db.all(`
    SELECT
      date_trunc('month', created_at) AS month,
      SUM(amount) AS revenue,
      COUNT(DISTINCT user_id) AS paying_users
    FROM orders
    WHERE created_at >= current_date - interval '${months} months'
      AND status = 'completed'
    GROUP BY 1
    ORDER BY 1
  `);
  return result;
}

// Use in an API endpoint
app.get("/api/analytics/revenue", async (req, res) => {
  const data = await getMonthlyRevenue(parseInt(req.query.months) || 12);
  res.json(data);
});
```

### Scheduled Transformations

```python
# scripts/daily_transform.py — Run as a cron job to update materialized tables
import duckdb

conn = duckdb.connect("md:analytics")

# Daily aggregation — rebuild summary tables
conn.sql("""
  CREATE OR REPLACE TABLE daily_kpis AS
  SELECT
    DATE(created_at) AS date,
    COUNT(DISTINCT user_id) AS dau,
    COUNT(DISTINCT CASE WHEN event_type = 'signup' THEN user_id END) AS signups,
    COUNT(DISTINCT CASE WHEN event_type = 'purchase' THEN user_id END) AS purchasers,
    SUM(CASE WHEN event_type = 'purchase' THEN amount ELSE 0 END) AS revenue,
    COUNT(*) AS total_events
  FROM events
  WHERE created_at >= current_date - interval '90 days'
  GROUP BY 1
  ORDER BY 1
""")

# Cohort analysis table
conn.sql("""
  CREATE OR REPLACE TABLE cohort_retention AS
  WITH first_seen AS (
    SELECT user_id, MIN(DATE(created_at)) AS cohort_date
    FROM events
    GROUP BY 1
  )
  SELECT
    f.cohort_date,
    DATE_DIFF('week', f.cohort_date, DATE(e.created_at)) AS week_number,
    COUNT(DISTINCT e.user_id) AS active_users
  FROM events e
  JOIN first_seen f ON e.user_id = f.user_id
  WHERE f.cohort_date >= current_date - interval '12 weeks'
  GROUP BY 1, 2
""")

print("✅ Daily transformations complete")
```

## Installation

```bash
# Python
pip install duckdb

# Node.js
npm install duckdb-async

# CLI (DuckDB CLI with MotherDuck support)
# Download from https://motherduck.com/docs/getting-started/
```


## Examples


### Example 1: Setting up an evaluation pipeline for a RAG application

**User request:**

```
I have a RAG chatbot that answers questions from our docs. Set up Motherduck to evaluate answer quality.
```

The agent creates an evaluation suite with appropriate metrics (faithfulness, relevance, answer correctness), configures test datasets from real user questions, runs baseline evaluations, and sets up CI integration so evaluations run on every prompt or retrieval change.

### Example 2: Comparing model performance across prompts

**User request:**

```
We're testing GPT-4o vs Claude on our customer support prompts. Set up a comparison with Motherduck.
```

The agent creates a structured experiment with the existing prompt set, configures both model providers, defines scoring criteria specific to customer support (accuracy, tone, completeness), runs the comparison, and generates a summary report with statistical significance indicators.


## Guidelines

1. **Hybrid execution** — MotherDuck pushes queries to the cloud OR runs locally depending on where data lives; leverage both
2. **Parquet for everything** — Store raw data as Parquet files on S3; query them directly without loading into tables
3. **Materialized tables for dashboards** — Pre-compute daily/weekly aggregates; don't recalculate on every dashboard load
4. **Share databases, not exports** — Instead of emailing CSVs, share a MotherDuck database; recipients get live, queryable data
5. **Use DuckDB locally for development** — Write and test queries locally with DuckDB; deploy the same SQL to MotherDuck
6. **Token management** — Store `MOTHERDUCK_TOKEN` in environment variables; never commit tokens to code
7. **Incremental loads** — For large datasets, append new data with `INSERT INTO ... WHERE NOT IN`; avoid full reloads
8. **Cost awareness** — MotherDuck charges by query compute; pre-aggregate expensive queries into summary tables