Snowflake

An MCP server for interacting with Snowflake databases.

A Model Context Protocol (MCP) server that provides access to Snowflake databases for any MCP-compatible client.

This server implements the Model Context Protocol to allow any MCP client to:

- Execute SQL queries on Snowflake databases
- Automatically handle database connection lifecycle (connect, reconnect on timeout, close)
- Handle query results and errors
- Perform database operations safely
- Connect using either password or key pair authentication

MCP is a standard protocol that allows applications to communicate with AI models and external services. It enables AI models to access tools and data sources beyond their training data, expanding their capabilities through a standardized communication interface. Key features include:

- Based on stdio communication (standard input/output)
- Structured tool definition and discovery
- Standardized tool call mechanism
- Structured results transmission

The Snowflake-MCP server consists of several key components:
- MCP Server- Central component that implements the MCP protocol and handles client requests
- Snowflake Connection Manager- Manages database connections, including creation, maintenance, and cleanup
- Query Processor- Executes SQL queries on Snowflake and processes the results
- Authentication Manager- Handles different authentication methods (password or private key)

The system works through the following communication flow:
- An MCP Client (such as Claude or other MCP-compatible application) sends a request to the MCP Server
- The MCP Server authenticates with Snowflake using credentials from the.envfile
- The MCP Server executes SQL queries on Snowflake
- Snowflake returns results to the MCP Server
- The MCP Server formats and sends the results back to the MCP Client

This architecture allows for seamless integration between AI applications and Snowflake databases while maintaining security and efficient connection management.

git clone https://github.com/davidamom/snowflake-mcp.git

Below is an example configuration for Claude Desktop, but this server works with any MCP-compatible client. Each client may have its own configuration method:

{ "mcpServers": { "snowflake": { "command": "C:\\Users\\YourUsername\\path\\to\\python.exe", "args": ["C:\\path\\to\\snowflake-mcp\\server.py"] } } }

- command: Full path to your Python interpreter. Please modify this according to your Python installation location.
- args: Full path to the server script. Please modify this according to where you cloned the repository.

Example paths for different operating systems:

{ "mcpServers": { "snowflake": { "command": "C:\\Users\\YourUsername\\anaconda3\\python.exe", "args": ["C:\\Path\\To\\snowflake-mcp\\server.py"] } } }

{ "mcpServers": { "snowflake": { "command": "/usr/bin/python3", "args": ["/path/to/snowflake-mcp/server.py"] } } }

Create a.envfile in the project root directory and add the following configuration:

# Snowflake Configuration - Basic Info SNOWFLAKE_USER=your_username # Your Snowflake username SNOWFLAKE_ACCOUNT=YourAccount.Region # Example: MyOrg.US-WEST-2 SNOWFLAKE_DATABASE=your_database # Your database SNOWFLAKE_WAREHOUSE=your_warehouse # Your warehouse SNOWFLAKE_ROLE=your_role # Your role # Authentication - Choose one method

This MCP server supports two authentication methods:

SNOWFLAKE_PASSWORD=your_password # Your Snowflake password

SNOWFLAKE_PRIVATE_KEY_FILE=/path/to/rsa_key.p8 # Path to private key file SNOWFLAKE_PRIVATE_KEY_PASSPHRASE=your_passphrase # Optional: passphrase if key is encrypted

For key pair authentication, you must first set up key pair authentication with Snowflake:

- Generate a key pair and register the public key with Snowflake
- Store the private key file securely on your machine
- Provide the full path to the private key file in the configuration

For instructions on setting up key pair authentication, refer toSnowflake documentation on key pair authentication.

{ "mcpServers": { "snowflake-docker": { "command": "docker", "args": [ "run", "-i", "snowflake-mcp" ], "env": { "SNOWFLAKE_USER": "your_username", "SNOWFLAKE_ACCOUNT": "your_account", "SNOWFLAKE_DATABASE": "your_database", "SNOWFLAKE_WAREHOUSE": "your_warehouse", "SNOWFLAKE_PASSWORD": "your_password", "SNOWFLAKE_ROLE": "your_role" } } } }

Note: The Docker implementation uses stdio for communication, so no ports need to be exposed.

If using key pair authentication with Docker, you'll need to mount your private key file:

docker run -i -v /path/to/your/key.p8:/app/rsa_key.p8:ro snowflake-mcp

And update your configuration accordingly:

{ "mcpServers": { "Snowflake-Docker": { "command": "docker", "args": [ "run", "-i", "-v", "/path/to/your/key.p8:/app/rsa_key.p8:ro", //optional "-v", "/path/to/export/dir/:/export/" "snowflake-mcp" ], "env": { "SNOWFLAKE_USER": "your_username", "SNOWFLAKE_ACCOUNT": "your_account", "SNOWFLAKE_DATABASE": "your_database", "SNOWFLAKE_WAREHOUSE": "your_warehouse", "SNOWFLAKE_ROLE": "your_role", "SNOWFLAKE_PRIVATE_KEY_FILE": "path_for_your_private_key", "SNOWFLAKE_PRIVATE_KEY_PASSPHRASE": "your_password_for_private_key" } } } }

- Secure Snowflake database access
- Flexible authentication (password or key pair authentication)
- Robust error handling and reporting
- Automatic connection management
- Query execution and result processing
- Compatible with any MCP-compliant client

The implementation consists of several key classes and modules:

- server.py- The main entry point containing the MCP server implementation.
- SnowflakeConnection- Class that handles all Snowflake database operations, including:

- Connection establishment and reconnection
- Query execution and transaction management
- Connection maintenance and cleanup

- Registers available tools with the MCP framework
- Handles tool call requests from clients
- Manages the lifecycle of connections

The connection lifecycle is carefully managed to ensure reliability:
- Initialization- Connections are created lazily when the first query is received
- Validation- Connection parameters are validated before attempting to connect
- Monitoring- Connections are regularly tested for validity
- Recovery- Automatic reconnection if the connection is lost or times out
- Cleanup- Proper resource release when the server shuts down

The server exposes the following tool to MCP clients:

-

execute_query- Executes a SQL query on Snowflake and returns the results

- Input: SQL query string
- Output: Query results in a structured format

export_to_csv- Executes a SQL query on Snowflake and returns the results

- Input: SQL query string
- Output: Num rows exported. File path of the output file

This implementation follows best practices for both MCP protocol implementation and Snowflake database interaction.

This project is licensed under theMIT License. See theLICENSEfile for details.

Query and analyze data with MotherDuck and local DuckDB