Snowflake Cortex AI Model Context Protocol (MCP) Server is a community-maintained MCP server project that exposes Snowflake capabilities as tools to AI clients that support MCP. The project is now deprecated and no longer maintained. The official recommendation is for all users to migrate to the official Snowflake MCP Server.
Purpose
The project is intended to enable AI clients (such as Claude Desktop, Cursor, fast-agent, Codex, etc.) to call Snowflake Cortex AI and related data services through the MCP protocol, enabling AI-driven data queries, object management, and SQL orchestration.
Main Capabilities
- Cortex Search: Query unstructured data in Snowflake, commonly used for retrieval-augmented generation (RAG) applications.
- Cortex Analyst: Query structured data through rich semantic modeling.
- Cortex Agent: Agent orchestration across structured and unstructured data retrieval.
- Object management: Perform basic operations such as create, delete, and update on common Snowflake objects.
- SQL execution: Run SQL generated by an LLM and governed by user-configured permissions.
- Semantic view queries: Discover and query Snowflake semantic views.
Usage
The server drives all tools through a configuration file (such as services/configuration.yaml). Multiple Cortex Agent, Search, and Analyst services can be added to the configuration file, and only services explicitly listed will be available as tools in the MCP client. Tool groups such as object management, query execution, and semantic views are enabled through Boolean switches in other_services. The sql_statement_permissions section is used to restrict the types of SQL expressions allowed to execute.
Connecting to Snowflake uses the Snowflake Python Connector and supports username/password, key pair, OAuth, SSO, multi-factor authentication (MFA), and other methods. Connection parameters can be passed through CLI arguments or environment variables, and follow the RBAC permissions of the specified role.
The server supports multiple transport mechanisms and canrun locally or be started by an MCP client, and also supports Docker and Docker Compose containerized deployment for remote access. The connection URL format is http://localhost:9000/snowflake-mcp（本地）或 http://<hostname>:<port>/snowflake-mcp (remote).
Use Cases
Suitable for scenarios where users want to execute Cortex AI queries, data retrieval, object management, and SQL operations on Snowflake through an MCP client, especially use cases such as RAG, joint queries over structured/unstructured data, and agent orchestration.
Limitations
- The project is deprecated, no longer maintained, and has no official support.
- Only Cortex Agent objects preconfigured in Snowflake are supported as tools.
- Object management tools cover only basic operations; for advanced management, using Snowsight directly is recommended.
- Not all Snowflake SQL commands are mapped in sqlglot; commands not captured require setting Unknown to True in order to pass validation.
- Deprecated CLI arguments (--account-identifier, --pat) and environment variables (SNOWFLAKE_PAT) will be removed in future versions.
- When using a Programmatic Access Token, secondary roles are not evaluated; a single role or any role must be selected.
Other details (such as specific charges, pricing, performance metrics, etc.) are not provided for now.
