Data warehouse · 1 skill
Snowflake MCP and Claude skills
Run SQL against Snowflake for your business metrics and for credit spend, failed tasks and slow queries from ACCOUNT_USAGE, and report what changed.
Connect Snowflake to your agent (MCP)
- Snowflake-managed MCP server. You create it in your account with
CREATE MCP SERVER ... FROM SPECIFICATION, listing tools such asSYSTEM_EXECUTE_SQL(withread_only: truein its config), Cortex Analyst, Cortex Search or a Cortex Agent. Its URL ishttps://<account_url>/api/v2/databases/<db>/schemas/<schema>/mcp-servers/<name>. Snowflake is in Claude's connector directory, and in Claude you can also add the URL as a custom connector with the client ID and secret from a Snowflake OAuth security integration, since Snowflake doesn't support dynamic client registration. In Claude Code, the simplest route is a programmatic access token for a least-privileged role:claude mcp add --transport http snowflake <url> --header "Authorization: Bearer <PAT>". Cursor takes the URL with an OAuth client ID and secret (docs). The older open-source Snowflake Labs MCP server is deprecated. - Snowflake CLI. With a connection named in
~/.snowflake/config.toml,snow sql -q "SELECT ..." -c reporting --format jsonreturns rows an agent can read. - Drivers or the SQL API with a service user and a programmatic access token; the skill reads
SNOWFLAKE_ACCOUNT,SNOWFLAKE_USER,SNOWFLAKE_PAT,SNOWFLAKE_ROLEandSNOWFLAKE_WAREHOUSE.
Create a reporting role with USAGE on one small warehouse and SELECT on the reporting schema, add the SNOWFLAKE.USAGE_VIEWER database role for the cost views, and grant the MCP server and its tools to that role only. Snowflake notes that tool annotations are hints, not controls: the role is what limits the agent.
1 Snowflake skill
What agents do with Snowflake
Agents query Snowflake with SQL, so a report can be anything your tables hold: revenue, signups, orders, pipeline. For the platform itself, the SNOWFLAKE.ACCOUNT_USAGE schema has the views cost and reliability reports need: WAREHOUSE_METERING_HISTORY (credits per warehouse per hour), METERING_DAILY_HISTORY, QUERY_HISTORY (TOTAL_ELAPSED_TIME, EXECUTION_STATUS, WAREHOUSE_NAME, USER_NAME), TASK_HISTORY for scheduled tasks and their errors, STORAGE_USAGE and LOGIN_HISTORY. These views lag by up to a few hours, so daily reports cover the previous full day. Cortex Analyst with semantic views lets an agent ask in plain language against definitions your data team maintains.
Data and finance teams schedule daily credit spend by warehouse against the trailing average, failed tasks, the slowest and most expensive queries of the week, storage growth, and a morning post of the core business metrics from a reporting schema.
Teams that use these
Your Snowflake numbers, on one board with everything else
Skills are free to use with any agent. Briefkin is where their results land: one board for every report, number and question your agents send, with a ping when something needs you.
Free during the open beta. No card.