Snowflake Cortex MCP Server gives an AI assistant access to Snowflake's Cortex AI capabilities - Cortex Search for querying unstructured data (RAG use cases), Cortex Analyst for querying structured data via semantic modeling and text-to-SQL, and Cortex Agent for orchestrating across both structured and unstructured data types.

Use it when you want to ask natural-language questions against Snowflake data - like "Show me the top selling brands by total sales quantity in TX for Books in 2003" - and have Cortex Analyst translate that into SQL against a semantic model, or use Cortex Search for RAG-style retrieval from unstructured data, or let Cortex Agent decide which data source to query. Do not use it without a Snowflake Personal Access Token (SNOWFLAKE_PAT) with the correct role configuration - PATs do not evaluate secondary roles, so make sure the token's primary role has the permissions your queries need.

• cortex_search: query unstructured data in Snowflake for RAG applications.
• cortex_analyst_text_to_sql: query structured data through advanced semantic modeling (text-to-SQL).
• cortex_agent: agent orchestrator for extracting structured and unstructured data across sources.
• sql_execution_tool: execute raw SQL queries directly.

Additional capabilities include a Payload Builder for dynamically constructing agent requests, streaming responses via SSE, and support for multiple Cortex Search and Analyst instances configured via environment variables.

yarn dev

Debug with the MCP Inspector: yarn inspector or npx @modelcontextprotocol/inspector tsx --env-file .env dist/mcp/MCP.js. Configure Claude Desktop:

{
  "mcpServers": {
    "Cortex Agent AI": {
      "command": "ABSOLUTE_PATH\\npx.cmd",
      "args": ["tsx", "--watch", "--env-file", "ABSOLUTE_PATH\\.env", "ABSOLUTE_PATH\\src\\mcp\\MCP.ts"]
    }
  }
}

Required: SNOWFLAKE_PAT. Optional: SEMANTIC_MODEL_VIEW (semantic model file for Text2SQL), VEHICLES_SEARCH_SERVICE (search service name). Also compatible with VS Code + GitHub Copilot and other MCP clients over stdio/sockets.

Data analysts and engineers using Snowflake Cortex who want an AI assistant to answer natural-language questions against both structured and unstructured Snowflake data without writing SQL or search queries by hand.