The four schema calls give an agent the orientation it needs before it writes SQL, and read_query is the only way data comes back — this build has no write path. append_insight is the unusual part: findings accumulate into a memo instead of scrolling out of the conversation.
A read path into Snowflake: walk from databases to schemas to tables to columns, run SELECT queries, and append findings to a memo the server keeps updated.
- Every database in the Snowflake instance
- The schemas inside a database you name
- The tables inside one database and schema
- Column names, types, nullability, defaults and comments for a table
- A SELECT query run against the warehouse, returned as rows
- A running memo of insights that updates as you append to it
A Snowflake account, with account, warehouse, user, password, role and database passed on the launch command. It runs through uvx.
One command plus a key — uvx mcp_snowflake_server_readonly, then supply credentials
