[ MCP SERVER SUMMARY ]
Snowflake MCP server summary
Snowflake's official Cortex MCP server (preferred over the deprecated Snowflake-Labs community server) lets AI agents use Cortex Search, Analyst, Agents, and governed SQL. Bifrost fronts Snowflake MCP so data tools are keyed, scoped, and audited for enterprise AI access.
| Property | Details |
|---|---|
| Company | Snowflake |
| Connection type | HTTP |
| Authentication | Snowflake account credentials / OAuth |
| Package or endpoint | Official Snowflake Cortex MCP |
| Capabilities | Cortex Search and Analyst, Cortex Agents, Governed SQL and object tools |
| Official documentation | Snowflake MCP documentation |
Before you start
- Snowflake account with Cortex MCP enabled
- Role with appropriate RBAC
What you can do with Snowflake
- Expose Cortex Analyst to analysts through Bifrost virtual keys
- Keep warehouse credentials out of IDE configs
- Enforce SQL statement permissions alongside Bifrost tool policies
[ HOW TO CONNECT ]
Connect Snowflake to Bifrost in three steps
Install Bifrost
Start the gateway locally or deploy in your VPC.
npx -y @maximhq/bifrost
Add Snowflake in MCP Gateway
Open MCP Gateway in the Bifrost dashboard, click New MCP Server, and choose HTTP. Or register via the API below.
• Prefer the official Snowflake Cortex MCP docs — the Snowflake-Labs/mcp GitHub repo is deprecated.
• Replace connection_string with your account’s MCP endpoint from Snowflake documentation.
Point clients at Bifrost /mcp
Configure Claude Desktop, Cursor, or Claude Code to use your Bifrost endpoint with a virtual key instead of connecting directly to Snowflake.
{
"mcpServers": {
"bifrost": {
"url": "http://localhost:8080/mcp",
"headers": {
"Authorization": "Bearer vk_your_virtual_key"
}
}
}
}[ EXAMPLE CONFIG ]
Register Snowflake via API
Add this MCP client with a POST to /api/mcp/client. See Connecting to MCP Servers for full options.
curl -X POST http://localhost:8080/api/mcp/client \
-H "Content-Type: application/json" \
-d '{
"name": "snowflake",
"connection_type": "http",
"connection_string": "https://<org>-<account>.snowflakecomputing.com/api/v2/mcp",
"auth_type": "oauth",
"tools_to_execute": [
"*"
]
}'[ WHY BIFROST ]
Run Snowflake through one governed gateway
Access control
Virtual keys, RBAC, and tool allowlists per team or agent.
Cost governance
Attribute MCP tool spend alongside LLM usage in one view.
Observability
Log every tool call with latency, server, and virtual key metadata.
Credential isolation
Upstream secrets stay in the gateway — not on every laptop.
[ FAQ ]
Common questions
What does the Snowflake MCP server do?
Snowflake exposes cortex search and analyst, cortex agents, governed sql and object tools to MCP-compatible AI clients. Bifrost connects to it as an upstream MCP client and exposes approved tools through one governed /mcp endpoint.
How does Snowflake authenticate when used with Bifrost?
Snowflake uses Snowflake account credentials / OAuth. Store credentials in Bifrost's environment or credential configuration so they stay at the gateway instead of being copied into every AI client.
Can Claude Desktop, Cursor, or Claude Code use Snowflake through Bifrost?
Yes. After registering Snowflake in Bifrost, point any HTTP MCP-compatible client at the Bifrost /mcp endpoint and authenticate with a virtual key. The client sees only the tools allowed for that key.
[ RELATED SERVERS ]
More MCP integrations
Apache Pinot
Query and explore real-time analytics data in Apache Pinot through StarTree’s MCP server.
Honeycomb
Give AI agents access to Honeycomb observability data for investigating production systems, traces, and telemetry.
Google Drive
Connect AI assistants to files and folders in Google Drive through Google Workspace MCP.