Creativity · MCP — server

MCP Snowflake Server

The MCP Snowflake server turns a Snowflake account into a callable toolset for LLM clients: run SQL, describe tables, list warehouses, and inspect query history. Snowflake maintains an official server in their developer-tools GitHub org, which pairs naturally with Cortex for LLM-driven analytics. Community builds add Snowpark Python execution and custom semantic-layer lookups.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio, http

Capabilities

Tools: run_query, describe_table, list_warehouses, get_query_history
Resources: database/schema/table metadata; Cortex function catalog
Auth: key-pair or OAuth against a Snowflake user/role

Install

uvx mcp_snowflake_server

Configuration

{
  "mcpServers": {
    "snowflake": {
      "command": "uvx",
      "args": ["mcp_snowflake_server"],
      "env": {
        "SNOWFLAKE_ACCOUNT": "xy12345.us-east-1",
        "SNOWFLAKE_USER": "svc_mcp",
        "SNOWFLAKE_PRIVATE_KEY_PATH": "/Users/you/.snowflake/rsa_key.p8",
        "SNOWFLAKE_ROLE": "ANALYST_RO",
        "SNOWFLAKE_WAREHOUSE": "ANALYST_WH"
      }
    }
  }
}

Frequently asked questions

Does Snowflake officially support MCP?

Yes — Snowflake Labs publishes an MCP server with tools for query execution, Cortex, and metadata browsing. It's the recommended starting point.

How should I authenticate?

Use key-pair auth with a dedicated service user and a read-only role (e.g. ANALYST_RO). Avoid user+password and avoid ACCOUNTADMIN for day-to-day LLM queries.

Can the server run Cortex functions like SUMMARIZE or EXTRACT_ANSWER?

Yes, through the run_query tool — Cortex functions are just SQL-callable. The official server may also expose dedicated tools for common Cortex functions depending on version.

Sources

Snowflake Labs MCP — accessed 2026-04-20
Snowflake Cortex LLM functions — accessed 2026-04-20