Summary

A warehouse you can ask about before you query it.

Schema browsing, the query itself and somewhere to keep the finding all live in one connection — and the destructive half stays absent until you start the server with the flag that adds it.

What it is

A server over a Snowflake connection. Reads are always available; the write and create-table tools exist only when the server was started with `--allow-write`. Alongside the tools it exposes per-table schema summaries and a running insights memo as MCP resources.

What you get

read_query runs a SELECT and returns rows as objects
Schema navigation without SQL: list_databases, list_schemas, list_tables, and describe_table for column names, types, nullability, defaults and comments
append_insight adds a finding to a memo the server keeps updated at memo://insights, so what you learned outlives the session
Per-table schema summaries are exposed as resources when prefetch is on, so context arrives without a round trip
write_query and create_table cover INSERT, UPDATE, DELETE and CREATE TABLE — present only when the server was launched with `--allow-write`

Requirements

A Snowflake account and connection details for it. The server is read-only unless you start it with `--allow-write`, which is the switch that turns on write_query and create_table.

Setup effort

One command plus a key — npx -y @smithery/cli install mcp_snowflake_server --client claude, then supply credentials