/snowflake-query

name: snowflake-query
description: >-
  Run SQL queries against Snowflake through the Managed MCP Server.
  Use for data retrieval, aggregations, joins, semi-structured data,
  time travel, and any explicit SQL the user requests.

Snowflake Query

Run SQL queries against Snowflake. Use this skill when the user asks to query data, run SQL, or retrieve specific results.

When to use

• User asks to query, select, or look at data
• User provides a specific SQL statement to run
• User asks about row counts, aggregations, or filtered results
• User needs joins, window functions, or CTEs

How to use

1. Identify the SQL. Use Snowflake dialect (ILIKE, TRY_CAST, FLATTEN, QUALIFY). 2. Send the query to the MCP server. With the default setup in this plugin's README, that's the `cortex_code_agent` tool. If the server exposes a dedicated SQL execution tool, use that instead. 3. Summarize results. For large sets, show a sample and total count.

Examples

**Simple count:**

SELECT COUNT(*) AS row_count FROM my_db.my_schema.orders;

**Top-N with aggregation:**

SELECT customer_name, SUM(amount) AS total_revenue
FROM orders
GROUP BY customer_name
ORDER BY total_revenue DESC
LIMIT 10;

**Join with filter:**

SELECT o.order_id, c.name, o.amount
FROM orders o
JOIN customers c ON o.customer_id = c.id
WHERE o.created_at >= DATEADD(day, -30, CURRENT_DATE())
ORDER BY o.amount DESC
LIMIT 20;

**Semi-structured (VARIANT/JSON):**

SELECT
  raw:event_type::STRING AS event_type,
  raw:user.name::STRING AS user_name,
  raw:timestamp::TIMESTAMP AS event_time
FROM events
WHERE raw:event_type::STRING ILIKE '%purchase%'
LIMIT 10;

**Flatten arrays:**

SELECT t.order_id, f.value:product_name::STRING AS product
FROM orders t,
LATERAL FLATTEN(input => t.items) f
LIMIT 20;

**Time travel:**

SELECT * FROM orders AT(TIMESTAMP => '2025-01-15 10:00:00'::TIMESTAMP) LIMIT 10;

**Window function with QUALIFY:**

SELECT customer_id, order_date, amount
FROM orders
QUALIFY ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY order_date DESC) = 1;

Important

• Default to read-only (SELECT). Never run DDL/DML unless explicitly requested.
• Always use LIMIT on exploratory queries.
• If a query fails, report the error and suggest a fix.
• MCP responses truncate at 250KB — aggregate or narrow your SELECT if results are large.