NEWS · CODING · #3769
Amazon Bedrock AgentCore outlines three OAuth patterns for authenticating to Snowflake-managed MCP/Cortex
The post describes three OAuth configurations for Amazon Bedrock AgentCore Identity/Gateway to authenticate AI agents to the Snowflake‑managed MCP server (which exposes Cortex Search and Cortex Analyst): Snowflake's built‑in OAuth (authorization code), three‑legged OAuth via a third‑party IdP (corporate SSO), and two‑legged client_credentials via a third‑party IdP for service users. It explains that AgentCore Gateway obtains a workload token (GetWorkloadAccessTokenForJWT), calls GetResourceOauth2Token, and that AgentCore Identity brokers, caches, and refreshes tokens and supports an on‑behalf‑of (TOKEN_EXCHANGE) flow; the patterns were tested with Okta and Microsoft Entra ID.
KEY POINTS
- The post describes three OAuth configurations for Amazon Bedrock AgentCore Identity/Gateway to authenticate AI agents to the Snowflake‑managed MCP server (which exposes Cortex Search and Cortex Analyst): Snowflake's built‑in OAuth (authorization code), three‑legged OAuth via a third‑party IdP (corporate SSO), and two‑legged client_credentials via a third‑party IdP for service users.
- It explains that AgentCore Gateway obtains a workload token (GetWorkloadAccessTokenForJWT), calls GetResourceOauth2Token, and that AgentCore Identity brokers, caches, and refreshes tokens and supports an on‑behalf‑of (TOKEN_EXCHANGE) flow; the patterns were tested with Okta and Microsoft Entra ID.
- This matters because it gives enterprises concrete, tested ways to provide per‑user and machine identity for AI agents calling Snowflake Cortex tools, avoiding broad service‑account permissions and custom auth plumbing.
WHY IT MATTERS
This matters because it gives enterprises concrete, tested ways to provide per‑user and machine identity for AI agents calling Snowflake Cortex tools, avoiding broad service‑account permissions and custom auth plumbing.