AWS tests three OAuth patterns for agents calling Snowflake Cortex via MCP

AWS ML Blog · rss · 2026-11-18

Background

Extending auth into the agentic layer typically means either overly broad service-account permissions or custom-built integrations. When an agent on Amazon Bedrock AgentCore calls the Snowflake-managed MCP server (exposing Cortex Search and Cortex Analyst) over MCP, the first question is how the agent authenticates.

Three patterns tested end-to-end

Key mechanics

The reference architecture adds a Strands agent on AgentCore Runtime, Bedrock Guardrails for PII redaction, and Knowledge Bases for RAG. Patterns 2/3 follow standard OAuth flows and extend to other OAuth-protected MCP servers.

Original post →

More from coding & agent

coding & agent channel →