Artificial intelligence news.
Straight from the original announcements.

AWS details three OAuth patterns for Snowflake Cortex AI agents

November 17, 2026, 21:25By the AI News Bites Editorial TeamSource: AWS
AWS details three OAuth patterns for Snowflake Cortex AI agents
Image: AWS

AWS has detailed three OAuth configurations for agents on Amazon Bedrock AgentCore that call Snowflake Cortex AI through Snowflake’s managed Model Context Protocol server. The patterns are designed to extend enterprise authentication and authorization controls into the agentic layer.

Key points
  • AWS tested three OAuth configurations for Snowflake Cortex AI through AgentCore.
  • AgentCore Identity stores, refreshes, and binds OAuth tokens to users or workloads.
  • Snowflake OAuth provides per-user access without an external identity provider.
  • Two-legged OAuth uses a shared service user for machine-to-machine workloads.
  • External identity provider patterns can extend to other OAuth-protected MCP servers.

Three ways to authenticate an agent

The AWS post examines Snowflake OAuth, three-legged OAuth through a third-party identity provider, and two-legged OAuth through a third-party identity provider. Amazon Bedrock AgentCore Identity supports these configurations through the authorization_code and client_credentials grant types.

In the Snowflake OAuth pattern, each user authenticates directly to Snowflake through an authorization code flow. The configuration does not require an external identity provider and is intended to preserve per-user identity. In the three-legged external identity provider pattern, users authenticate through corporate single sign-on, while Snowflake trusts identity provider-issued JSON Web Tokens through an External OAuth security integration.

The two-legged pattern uses the client credentials grant. A shared service user authenticates through the identity provider, with no user interaction or consent prompt. AWS presents this approach as suitable for machine-to-machine workloads.

AgentCore Identity brokers and refreshes credentials

Amazon Bedrock AgentCore Gateway connects agents to external tool providers through MCP server targets. In the described architecture, its target is the Snowflake-managed MCP server, which exposes Cortex Search and Cortex Analyst as MCP tools.

Amazon Bedrock AgentCore Identity acts as the credential broker for the Gateway’s outbound authentication. The Gateway first obtains a workload token using GetWorkloadAccessTokenForJWT, then requests an OAuth token with GetResourceOauth2Token from the configured provider. AgentCore Identity manages the token vault, token refresh, and user-to-token binding, and automatically fetches, caches, and refreshes tokens.

The broader reference architecture includes a user-facing AI application, a Strands agent on Amazon Bedrock AgentCore Runtime, Amazon Bedrock for reasoning, Amazon Bedrock Guardrails for redacting personally identifiable information, and Amazon Bedrock Knowledge Bases over policy documents in Amazon Simple Storage Service. The agent reaches Snowflake Cortex AI through AgentCore Gateway and the managed MCP server.

Choosing between per-user and shared access

AWS says the Snowflake OAuth and external identity provider three-legged patterns attribute Snowflake activity to the real user and their assigned role. In the Snowflake-specific pattern, subsequent calls run as that user and Snowflake’s QUERY_HISTORY records the user’s real username. Users consent to a Snowflake role, while Snowflake settings can narrow advertised role scopes and restrict which roles may be approved.

The two-legged pattern instead uses a shared Snowflake service user. It removes user interaction, but does not provide the same per-user identity described for the three-legged configurations. AgentCore Identity also supports an on-behalf-of grant, TOKEN_EXCHANGE, which AWS says can deliver per-user identity for unattended workloads without a consent prompt.

Patterns using an external identity provider follow standard OAuth flows and can extend to other OAuth-protected MCP servers, although AWS says token claims, scopes, and audience settings should be validated for each software-as-a-service provider. The Snowflake-specific configuration relies on Snowflake’s built-in OAuth authorization server and related Snowflake settings.

Related news

AWS lets AI agents buy x402 products through Marketplace billing
Products and toolsNovember 20, 2026, 00:02

AWS lets AI agents buy x402 products through Marketplace billing

AWS has added AWS Billing to Amazon Bedrock AgentCore payments, allowing AI agents to discover and purchase x402 enabled software, APIs, data and content through AWS…

AWS upgrades Bedrock AgentCore with Strands defaults and lifecycle hooks
For developersNovember 19, 2026, 21:01

AWS upgrades Bedrock AgentCore with Strands defaults and lifecycle hooks

Amazon Bedrock AgentCore harness now uses AWS’s open-source Strands harness as its foundation, adding managed defaults, built-in tools and lifecycle hooks for…

ThoughtSpot Brings Agentic Analytics Into Customers’ Own Cloud Accounts
Products and toolsNovember 18, 2026, 22:57

ThoughtSpot Brings Agentic Analytics Into Customers’ Own Cloud Accounts

ThoughtSpot has introduced Customer Hosted, a deployment option that places its full analytics platform, including Spotter and its governed semantic layer, inside a…