Snowflake
Arcade Optimized
Tools to query and explore Snowflake with per-user, per-role access
Author:Arcade
Version:
0.1.0Auth:User authorization
7tools
7require secrets
Snowflake Toolkit
Arcade's Snowflake toolkit provides read-only, per-user access to Snowflake — enabling agents and apps to explore schema structure and execute SELECT queries under each caller's own Snowflake identity and role.
Capabilities
- Identity & access inspection: Retrieve the active Snowflake user, role, and default warehouse; all other tools operate under this identity, so two users issuing identical calls can legitimately receive different results.
- Schema discovery (databases → schemas → tables → columns): Progressively explore the object hierarchy — list accessible databases, then schemas within a database, then tables/views within a schema, then column definitions for a specific table or view — filtered server-side to only what the caller's role is granted.
- Warehouse discovery: List warehouses the caller's role can use (smallest = cheapest); needed when no default warehouse is configured or a different size is preferred.
- Safe SELECT execution: Run arbitrary read-only SQL; writes, DDL, role changes, and data movement are rejected before reaching Snowflake. Supports
limit/offsetpagination parameters; returns column names, rows, and atruncatedflag.
OAuth
This toolkit uses OAuth 2.0 to authenticate each user with Snowflake. Arcade manages the OAuth flow; refer to the Arcade auth-provider documentation for setup details. (Provider-specific docs link unavailable — provider ID not supplied.)
Secrets
SNOWFLAKE_ACCOUNT_SUBDOMAIN— The account identifier subdomain for your Snowflake account (e.g.,xy12345.us-east-1or an organization-format identifier likemyorg-myaccount). This is the hostname prefix used to construct your Snowflake account URL (<subdomain>.snowflakecomputing.com). You can find it in the Snowflake UI under Admin → Accounts, or by inspecting the URL you use to log in. See Snowflake account identifier docs for format details.
Configure secrets in the Arcade secrets dashboard. For guidance on referencing secrets in tools, see the Arcade secrets docs.
Available tools(7)
7 of 7 tools
Operations
Behavior
| Tool name | Description | Secrets | |
|---|---|---|---|
List the databases your role can access, newest-created first.
Snowflake returns only databases your current role is granted, so this is
already a per-user answer. Take a `name` and use it to list that database's
schemas. Use `name_pattern` to filter server-side rather than paging the
whole list. | 1 | ||
List the schemas in a database that your role can access.
List the databases first to get a valid `database`. Then take a schema
`name` and list that schema's tables. | 1 | ||
List the tables and views in a schema that your role can query.
List the schemas first to get a valid `database` and `schema_name`, then
load a table's columns before querying it.
Each row's `kind` is TABLE, VIEW, MATERIALIZED VIEW, or SEMANTIC VIEW. A
SEMANTIC VIEW cannot be read with an ordinary SELECT: it needs Snowflake's
SEMANTIC_VIEW(...) syntax, which this toolkit does not support, so skip
those when choosing something to query. | 1 | ||
List the warehouses your role can run queries on.
An account usually has several, and Snowflake returns only the ones your
current role is granted, so this is already a per-user answer.
You only need this when a query has no warehouse to run on, either because
you have no default or because you want a different one. `size` is the cost
signal: prefer the smallest warehouse that fits the query. A SUSPENDED
warehouse is still usable, since Snowflake resumes it on use by default. | 1 | ||
Run a read-only SELECT query and return the rows.
Only SELECT is allowed. Writes, DDL, role changes, and data movement are
rejected before the query reaches Snowflake, so this tool cannot modify
anything even if your Snowflake role would permit it.
The query runs under your own Snowflake identity, so your role and
Snowflake's row-access and masking policies decide what you can read. Two
users running identical SQL can legitimately get different results.
Load the table's columns before calling this, and name the columns you want
rather than selecting everything. `rows` is positional and
aligned to `columns`. `truncated` is true when more rows were available than
were returned.
Do not put LIMIT, OFFSET, or SQL comments in any clause; use the `limit` and
`offset` parameters, which are applied for you. | 1 | ||
Get a table or view's columns, so you can select exact fields.
Load this before running a query, and select named columns rather than
everything. Works on views as well as tables. A not-found error usually
means the object is not granted to your role; re-check the table listing. | 1 | ||
Return the Snowflake identity every other tool uses, and the warehouse.
The role is the important field. It is what decides which objects the other
tools can see and which rows and cells come back, so two people calling the
same tool with the same arguments can legitimately get different data.
Pass `current_warehouse` straight through as the warehouse argument when
running a query. If it comes back empty this user has no default warehouse,
and a query must name one. The warehouse discovery tool lists the ones this
role may use, cheapest first by `size`.
This is the only place identity is reported. It runs no caller-supplied SQL,
so what it returns is always the identity the other tools will run under. | 1 |
Last updated on