MCP 服务器

Snowflake Developer MCP Server

Enables AI assistants to perform comprehensive Snowflake database operations including DDL, DML, and warehouse management. It allows users to query data, manage database objects, and configure permissions using natural language commands.

README

Snowflake Developer MCP Server 🚀

A powerful Model Context Protocol (MCP) server that provides comprehensive Snowflake database operations, Cortex AI services, and data management tools for AI assistants like Claude.

🌟 Features

🔧 DDL Operations: Create and manage databases, schemas, tables, and other database objects
📊 DML Operations: Insert, update, delete, and query data with full SQL support
⚙️ Snowflake Operations: Manage warehouses, grants, roles, and show database objects
🔒 Secure Authentication: Support for passwords and Programmatic Access Tokens (PAT)
🎯 Simple Connection Pattern: Per-operation connections for reliability and simplicity

🚀 Quick Start

Prerequisites

Python 3.11+
UV package manager (install from https://github.com/astral-sh/uv)
Node.js and npm (for MCP inspector)
Snowflake account with appropriate permissions
Snowflake credentials (account identifier, username, password/PAT)

Installation

Clone the repository

git clone https://github.com/mcp-tg/snowflake-developer.git
cd snowflake-developer

Set up environment

# Copy environment template
cp .env.example .env

# Edit .env with your Snowflake credentials
# Required: SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, SNOWFLAKE_PAT (or SNOWFLAKE_PASSWORD)

Install UV (if not already installed)

# On macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# On Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

🧪 Testing with MCP Inspector

The easiest way to test your setup is using the MCP Inspector:

# Run the development inspector script
./dev-inspector.sh

This will:

✅ Create a virtual environment (if needed)
✅ Install all dependencies via UV
✅ Load your Snowflake credentials from .env
✅ Start the MCP Inspector web interface
✅ Open your browser to test tools interactively

Note: The script automatically handles UV package installation, so you don't need to manually install dependencies.

First Test: Verify Connection

In the Inspector, go to the Tools tab
Find test_snowflake_connection and click Run
You should see your account details and confirmation that the connection works

🔌 Integration with AI Assistants

Claude Desktop

Option 1: Direct from GitHub (no local clone needed)

{
  "mcpServers": {
    "snowflake-developer": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/mcp-tg/snowflake-developer.git",
        "main.py"
      ],
      "env": {
        "SNOWFLAKE_ACCOUNT": "your-account",
        "SNOWFLAKE_USER": "your-username",
        "SNOWFLAKE_PAT": "your-pat-token"
      }
    }
  }
}

Option 2: Local installation

{
  "mcpServers": {
    "snowflake-developer": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/path/to/snowflake-developer",
        "python",
        "main.py"
      ],
      "env": {
        "SNOWFLAKE_ACCOUNT": "your-account",
        "SNOWFLAKE_USER": "your-username",
        "SNOWFLAKE_PAT": "your-pat-token"
      }
    }
  }
}

Setup Instructions:

Clone the repository: git clone https://github.com/mcp-tg/snowflake-developer.git
Create the Claude Desktop config file: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
Add the configuration above, replacing /path/to/snowflake-developer with your actual path
Replace credential placeholders with your actual Snowflake credentials
Restart Claude Desktop

Cursor

Note: Cursor doesn't support environment variables in MCP configuration. You'll need to use the local installation option or set environment variables globally on your system.

Option 1: Direct from GitHub (requires global env vars)

{
  "mcpServers": {
    "snowflake-developer": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/mcp-tg/snowflake-developer.git",
        "main.py"
      ]
    }
  }
}

Requires setting SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, and SNOWFLAKE_PAT as system environment variables.

Option 2: Local installation (recommended for Cursor)

{
  "mcpServers": {
    "snowflake-developer": {
      "command": "uv",
      "args": ["run", "/path/to/snowflake-developer/main.py"]
    }
  }
}

Use a local .env file in the project directory with your credentials.

📚 Available Tools (22 Total)

🔧 DDL Tools (8 Tools)

Tools for managing database structure:

Tool	Description	Example in Inspector	Natural Language Query
`alter_database`	Rename databases	database_name: `OLD_DB`<br>new_name: `NEW_DB`	"Rename database OLD_DB to NEW_DB"
`alter_schema`	Rename or move schemas	schema_name: `TEST_DB.OLD_SCHEMA`<br>new_name: `NEW_SCHEMA`	"Rename OLD_SCHEMA to NEW_SCHEMA in TEST_DB"
`alter_table`	Modify table structure	table_name: `TEST_DB.PUBLIC.USERS`<br>alter_type: `ADD`<br>column_name: `created_at`<br>data_type: `TIMESTAMP`	"Add a created_at timestamp column to TEST_DB.PUBLIC.USERS table"
`create_database`	Create a new database	database_name: `TEST_DB`	"Create a new database called TEST_DB"
`create_schema`	Create a schema in a database	database_name: `TEST_DB`<br>schema_name: `ANALYTICS`	"Create a schema named ANALYTICS in TEST_DB database"
`create_table`	Create a table with columns	database_name: `TEST_DB`<br>schema_name: `PUBLIC`<br>table_name: `USERS`<br>columns: `[{"name": "id", "type": "INT"}, {"name": "email", "type": "VARCHAR(255)"}]`	"Create a USERS table in TEST_DB.PUBLIC with id as INT and email as VARCHAR(255)"
`drop_database_object`	Drop any database object	object_type: `TABLE`<br>object_name: `TEST_DB.PUBLIC.OLD_TABLE`	"Drop the table TEST_DB.PUBLIC.OLD_TABLE"
`execute_ddl_statement`	Run custom DDL SQL	ddl_statement: `CREATE VIEW TEST_DB.PUBLIC.ACTIVE_USERS AS SELECT * FROM TEST_DB.PUBLIC.USERS WHERE status = 'active'`	"Create a view called ACTIVE_USERS that shows only active users"

📊 DML Tools (6 Tools)

Tools for working with data:

Tool	Description	Example in Inspector	Natural Language Query
`delete_data`	Delete rows from a table	table_name: `TEST_DB.PUBLIC.USERS`<br>where_clause: `status = 'deleted'`	"Delete all users with status 'deleted'"
`execute_dml_statement`	Run custom DML SQL	dml_statement: `UPDATE TEST_DB.PUBLIC.USERS SET last_login = CURRENT_TIMESTAMP() WHERE id = 1`	"Update the last login timestamp for user with id 1"
`insert_data`	Insert rows into a table	table_name: `TEST_DB.PUBLIC.USERS`<br>data: `{"id": 1, "email": "john@example.com", "name": "John Doe"}`	"Insert a new user with id 1, email john@example.com, and name John Doe into the USERS table"
`merge_data`	Synchronize data between tables	target_table: `TEST_DB.PUBLIC.USERS`<br>source_table: `TEST_DB.STAGING.NEW_USERS`<br>merge_condition: `target.id = source.id`<br>match_actions: `[{"action": "UPDATE", "columns": ["email", "name"], "values": ["source.email", "source.name"]}]`<br>not_match_actions: `[{"action": "INSERT", "columns": ["id", "email", "name"], "values": ["source.id", "source.email", "source.name"]}]`	"Merge new users from staging table into production users table, updating existing records and inserting new ones"
`query_data`	Query data from tables	table_name: `TEST_DB.PUBLIC.USERS`<br>columns: `["id", "email", "name"]`<br>where_clause: `status = 'active'`<br>limit: `10`	"Show me the first 10 active users with their id, email, and name"
`update_data`	Update existing rows	table_name: `TEST_DB.PUBLIC.USERS`<br>data: `{"status": "inactive"}`<br>where_clause: `last_login < '2023-01-01'`	"Set status to inactive for all users who haven't logged in since January 2023"

⚙️ Snowflake Operations Tools (8 Tools)

Tools for Snowflake-specific operations:

Tool	Description	Example in Inspector	Natural Language Query
`alter_warehouse`	Modify warehouse settings	warehouse_name: `COMPUTE_WH`<br>warehouse_size: `MEDIUM`<br>auto_suspend: `300`	"Change COMPUTE_WH to MEDIUM size and auto-suspend after 5 minutes"
`describe_database_object`	Get object details	object_name: `TEST_DB.PUBLIC.USERS`	"Describe the structure of TEST_DB.PUBLIC.USERS table"
`execute_sql_query`	Run any SQL query	query: `SELECT CURRENT_USER(), CURRENT_WAREHOUSE()`	"Show me my current user and warehouse"
`grant_privileges`	Grant permissions	privileges: `["SELECT", "INSERT"]`<br>on_type: `TABLE`<br>on_name: `TEST_DB.PUBLIC.USERS`<br>to_type: `ROLE`<br>to_name: `ANALYST_ROLE`	"Grant SELECT and INSERT on TEST_DB.PUBLIC.USERS table to ANALYST_ROLE"
`revoke_privileges`	Revoke permissions	privileges: `["SELECT"]`<br>on_type: `TABLE`<br>on_name: `TEST_DB.PUBLIC.USERS`<br>from_type: `ROLE`<br>from_name: `ANALYST_ROLE`	"Revoke SELECT on TEST_DB.PUBLIC.USERS table from ANALYST_ROLE"
`set_context`	Set database/schema/warehouse/role	context_type: `DATABASE`<br>context_name: `TEST_DB`	"Use TEST_DB as the current database"
`show_database_objects`	List database objects	object_type: `DATABASES`	"Show me all databases"
`test_snowflake_connection`	Test connection to Snowflake	(no parameters)	"Test my Snowflake connection"

🏗️ Architecture

The server uses a simple per-operation connection pattern:

Each tool/resource call creates a fresh Snowflake connection
Connections are automatically closed after each operation
No connection pooling or persistence required
Credentials are read from environment variables

🛡️ Security Best Practices

Use Programmatic Access Tokens (PAT) instead of passwords when possible
Never commit .env files to version control
Use least-privilege roles for your Snowflake user
Rotate credentials regularly
Consider using external secret management for production

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🆘 Troubleshooting

Connection Issues

Verify your account identifier format
Check that your user has appropriate permissions
Ensure your PAT token hasn't expired
Test network connectivity to Snowflake

Tool Errors

Check the error message in the Inspector console
Verify required parameters are provided
Ensure database objects exist before referencing them
Check SQL syntax for custom statements

🚀 FastMCP Framework

This MCP server is built using FastMCP, a modern Python framework that simplifies building Model Context Protocol servers. FastMCP provides:

Why FastMCP?

🎯 Simple API: Decorator-based tool and resource registration
⚡ High Performance: Async/await support with efficient message handling
🔧 Type Safety: Full TypeScript-style type hints and validation
📝 Auto Documentation: Automatic tool/resource documentation generation
🛡️ Error Handling: Built-in exception handling and response formatting
🔌 MCP Compliance: Full compatibility with MCP protocol specification

FastMCP vs Traditional MCP

# Traditional MCP server setup
class MyMCPServer:
    def __init__(self):
        self.tools = {}
    
    def register_tool(self, name, handler, schema):
        # Manual registration and validation
        pass

# FastMCP - Clean and Simple
from fastmcp import FastMCP

mcp = FastMCP("MyServer")

@mcp.tool()
def my_tool(param: str) -> str:
    """Tool with automatic type validation and documentation."""
    return f"Result: {param}"

@mcp.resource("my://resource/{id}")
async def my_resource(id: str, ctx: Context) -> dict:
    """Resource with built-in async support and context."""
    return {"data": f"Resource {id}"}

Key FastMCP Features Used

Decorator Registration: Tools are registered using simple decorators
Type Validation: Automatic parameter validation using Python type hints
Context Management: Built-in context for progress reporting and logging
Resource Patterns: URI template matching for dynamic resource endpoints
Error Handling: Automatic exception catching and standardized error responses

FastMCP Installation

# Install FastMCP
pip install fastmcp

# Or with UV (recommended)
uv add fastmcp