• • • •

Test
Lint
Meta
Security
Best Practices
Documentation

A Model Context Protocol (MCP) server / MCP server that connects AI assistants to Snowflake — enabling SQL queries, schema exploration, and data insights directly from your LLM client.

Highlights:

• Multiple authentication methods: password, key-pair, external browser, OAuth 2.0 (client credentials & bearer token), TOML connection files
• TOML multi-connection config — manage production, staging, and development environments in one file
• Write-safety guard — write operations are disabled by default and must be explicitly enabled
• Exclusion patterns — filter out databases, schemas, or tables from discovery
• --exclude-json-results flag — reduces LLM context window usage
• Selective tool exclusion via --exclude_tools
• Prefetch mode — pre-load table schema as MCP resources
• Docker support with hardened image (DHI, nonroot user, no shell in runtime)

• Snowflake MCP Server NSP
  • Table of Contents
  • Quick Start
    • Claude Code
    • Visual Studio Code (VSCode)
    • OpenCode
  • Components
    • Resources
    • Tools
      • Query Tools
      • Schema Tools
      • Analysis Tools
  • Authentication
    • Password
    • Key-Pair
    • External Browser
    • OAuth 2.0 Client Credentials
    • OAuth Bearer Token
    • TOML Connection File (Recommended)
  • Installation
    • Via UVX
    • Via Docker Hub
  • Configuration Reference
  • Exclusion Patterns
  • License
  • Fork and Attribution

The fastest way to try it — using uvx with a TOML connection file:

# 1. Create a connections file
cat > ~/snowflake_connections.toml << 'EOF'
[myconn]
account = "your_account"
user = "your_user"
password = "your_password"
warehouse = "COMPUTE_WH"
database = "MY_DB"
schema = "PUBLIC"
role = "MYROLE"
EOF

# 2. Run the server
uvx --python=3.13 --from mcp-snowflake-server-nsp mcp_snowflake_server \
  --connections-file ~/snowflake_connections.toml \
  --connection-name myconn

Add to your MCP client config (e.g. claude_desktop_config.json) using snowflake_connections.toml:

"mcpServers": {
  "snowflake": {
    "command": "uvx",
    "args": [
      "--python=3.13",
      "--from", "mcp-snowflake-server-nsp",
      "mcp_snowflake_server",
      "--connections-file", "/absolute/path/to/snowflake_connections.toml",
      "--connection-name", "myconn"
    ]
  }
}

uvx —

Docker —

Or add manually to your MCP client config (e.g. .vscode/mcp.json) using .env file (see Authentication):

"snowflake": {
      // Snowflake MCP server
      "type": "stdio",
      "command": "uvx",
      "args": [
        "--from", "mcp-snowflake-server-nsp",
        "--python=3.13",
        "mcp_snowflake_server"
      ],
      "envFile": "${workspaceFolder}/.env"
    }

Add to your MCP client config (e.g. opencode.jsonc) with .env file (see Authentication):

"snowflake": {
  "type": "local",
  "command": [
    "uvx",
    "--from",
    "mcp-snowflake-server-nsp",
    "--python=3.13",
    "mcp_snowflake_server",
  ],
  "enabled": true,
  "timeout": 300000,
}

Set credentials via environment variables or CLI flags (see Configuration Reference):

SNOWFLAKE_USER="user@example.com"
SNOWFLAKE_ACCOUNT="myaccount"
SNOWFLAKE_AUTHENTICATOR="snowflake"
SNOWFLAKE_PASSWORD="secret"
SNOWFLAKE_WAREHOUSE="COMPUTE_WH"
SNOWFLAKE_DATABASE="MY_DB"
SNOWFLAKE_SCHEMA="PUBLIC"
SNOWFLAKE_ROLE="MYROLE"

Both RSA (RS256) and ECDSA (ES256, ES384, ES512) private keys are supported (requires snowflake-connector-python ≥ 4.5.0 for ECDSA).

SNOWFLAKE_USER="user@example.com"
SNOWFLAKE_ACCOUNT="myaccount"
SNOWFLAKE_AUTHENTICATOR="snowflake_jwt"
SNOWFLAKE_PRIVATE_KEY_FILE="/absolute/path/to/key.p8"
SNOWFLAKE_PRIVATE_KEY_FILE_PWD="passphrase"  # Optional — only if key is encrypted
SNOWFLAKE_WAREHOUSE="COMPUTE_WH"
SNOWFLAKE_DATABASE="MY_DB"
SNOWFLAKE_SCHEMA="PUBLIC"
SNOWFLAKE_ROLE="MYROLE"

Or via CLI: --private_key_file /path/to/key.p8 --private_key_file_pwd passphrase

SNOWFLAKE_AUTHENTICATOR="externalbrowser"

Or in a TOML connection entry: authenticator = "externalbrowser"

Use the OAuth 2.0 client credentials flow to authenticate with a client ID and secret (no user interaction required):

SNOWFLAKE_AUTHENTICATOR="oauth_client_credentials"
SNOWFLAKE_ACCOUNT="myaccount"
SNOWFLAKE_OAUTH_CLIENT_ID="your_client_id"
SNOWFLAKE_OAUTH_CLIENT_SECRET="your_client_secret"
SNOWFLAKE_OAUTH_TOKEN_REQUEST_URL="https://your-idp.example.com/oauth/token"
SNOWFLAKE_OAUTH_SCOPE="session:role:MY_ROLE"  # Optional
SNOWFLAKE_WAREHOUSE="COMPUTE_WH"
SNOWFLAKE_DATABASE="MY_DB"
SNOWFLAKE_SCHEMA="PUBLIC"
SNOWFLAKE_ROLE="MYROLE"

Use a pre-fetched OAuth bearer token:

SNOWFLAKE_AUTHENTICATOR="oauth"
SNOWFLAKE_ACCOUNT="myaccount"
SNOWFLAKE_TOKEN="eyJhbGciOiJSUzI1NiJ9..."
SNOWFLAKE_WAREHOUSE="COMPUTE_WH"
SNOWFLAKE_DATABASE="MY_DB"
SNOWFLAKE_SCHEMA="PUBLIC"
SNOWFLAKE_ROLE="MYROLE"

Manage multiple environments in a single file. See example_connections.toml for a full template.

[production]
account = "your_account"
user = "your_user"
password = "your_password"
authenticator = "snowflake"
warehouse = "COMPUTE_WH"
database = "PROD_DB"
schema = "PUBLIC"
role = "ACCOUNTADMIN"

[development]
account = "your_account"
user = "dev_user"
authenticator = "externalbrowser"
warehouse = "DEV_WH"
database = "DEV_DB"
schema = "PUBLIC"
role = "DEVELOPER"

[reporting]
account = "your_account"
user = "reporting_user"
authenticator = "snowflake_jwt"
private_key_file = "/path/to/private_key.pem"
private_key_file_pwd = "passphrase"  # Optional
warehouse = "REPORTING_WH"
database = "REPORTING_DB"
schema = "REPORTS"
role = "REPORTING_ROLE"

[analytics_oauth]
account = "your_account"
authenticator = "oauth_client_credentials"
oauth_client_id = "your_client_id"
oauth_client_secret = "your_client_secret"
oauth_token_request_url = "https://your-idp.example.com/oauth/token"
oauth_scope = "session:role:ANALYTICS_ROLE"  # Optional
warehouse = "ANALYTICS_WH"
database = "ANALYTICS_DB"
schema = "PUBLIC"
role = "ANALYTICS_ROLE"

Pass the file with --connections-file and select a profile with --connection-name. Both flags are required together.

The package is published on PyPI as mcp-snowflake-server-nsp.

Contributing or running from source? See CONTRIBUTING.md for local development setup, test commands, formatting, and building the Docker image from source.

"mcpServers": {
  "snowflake_production": {
    "command": "uvx",
    "args": [
      "--python=3.13",
      "--from", "mcp-snowflake-server-nsp",
      "mcp_snowflake_server",
      "--connections-file", "/path/to/snowflake_connections.toml",
      "--connection-name", "production"
      // Optional flags — see Configuration Reference
    ]
  },
  "snowflake_staging": {
    "command": "uvx",
    "args": [
      "--python=3.13",
      "--from", "mcp-snowflake-server-nsp",
      "mcp_snowflake_server",
      "--connections-file", "/path/to/snowflake_connections.toml",
      "--connection-name", "staging"
    ]
  }
}

"mcpServers": {
  "snowflake": {
    "command": "uvx",
    "args": [
      "--python=3.13",
      "--from", "mcp-snowflake-server-nsp",
      "mcp_snowflake_server",
      "--account", "your_account",
      "--warehouse", "your_warehouse",
      "--user", "your_user",
      "--password", "your_password",
      "--role", "your_role",
      "--database", "your_database",
      "--schema", "your_schema"
      // Optional: "--private_key_file", "/absolute/path/key.p8"
      // Optional: "--private_key_file_pwd", "passphrase"
      // Optional flags — see Configuration Reference
    ]
  }
}

The image is published on Docker Hub — no build step required:

docker pull nsphung/mcp-snowflake-server-nsp

Note: -i (--interactive) is required to keep stdin open for the MCP stdio transport. Do not use -d (detach).

"mcpServers": {
  "snowflake": {
    "command": "docker",
    "args": [
      "run", "--rm", "-i",
      "--env-file", "/absolute/path/to/.env",
      "nsphung/mcp-snowflake-server-nsp"
    ]
  }
}

"mcpServers": {
  "snowflake": {
    "command": "docker",
    "args": [
      "run", "--rm", "-i",
      "-v", "/path/to/snowflake_connections.toml:/app/snowflake_connections.toml:ro",
      "nsphung/mcp-snowflake-server-nsp",
      "--connections-file", "/app/snowflake_connections.toml",
      "--connection-name", "production"
    ]
  }
}

"snowflake": {
  "type": "stdio",
  "command": "docker",
  "args": [
    "run", "--rm", "-i",
    "nsphung/mcp-snowflake-server-nsp"
  ],
  "envFile": "${workspaceFolder}/.env"
}

"snowflake": {
  "type": "stdio",
  "command": "docker",
  "args": [
    "run", "--rm", "-i",
    "-v", "/path/to/snowflake_connections.toml:/app/snowflake_connections.toml:ro",
    "nsphung/mcp-snowflake-server-nsp",
    "--connections-file", "/app/snowflake_connections.toml",
    "--connection-name", "production"
  ]
}

"snowflake": {
  "type": "local",
  "command": [
    "docker", "run", "--rm", "-i",
    "--env-file", "/absolute/path/to/.env",
    "nsphung/mcp-snowflake-server-nsp"
  ],
  "enabled": true,
  "timeout": 300000
}

All connection parameters can also be set as environment variables (SNOWFLAKE_<PARAM_UPPER>).

Edit runtime_config.json to exclude databases, schemas, or tables from all discovery tools. Patterns are matched case-insensitively as substrings.

{
  "exclude_patterns": {
    "databases": ["temp"],
    "schemas": ["temp", "information_schema"],
    "tables": ["temp"]
  }
}

The server loads this file automatically at startup from the working directory.

This project is licensed under the MIT License. See the LICENSE file for the full text.

This repository is a fork of isaacwasserman/mcp-snowflake-server.

• Upstream authors and contributors retain copyright for their contributions.
• Fork-specific changes are maintained by nsphung.
• A summary of notable modifications is tracked in NOTICE.