AWS Introduces OAuth Support for Model Context Protocol Server
AWS has launched OAuth support for its MCP Server, enabling browser-based authentication for AI agents using existing IAM and Identity Center credentials.
AWS now supports industry-standard OAuth for its Model Context Protocol server, streamlining secure AI agent integration.
- AWS now supports OAuth-based authentication for its Model Context Protocol (MCP) Server, allowing agents to use existing AWS credentials via a browser.
- The update includes new security tools such as dynamic client registration, token introspection, and global condition keys for OAuth.
- The system is compatible with IAM federation, AWS IAM Identity Center, and standard IAM or root users without granting agents extra permissions.
OAuth Integration for AI Agent Connectivity
Amazon Web Services (AWS) has announced the introduction of OAuth support for the AWS MCP Server. This update allows developers to connect AI agents to AWS resources using the same authentication methods and credentials employed for the AWS Management Console or the AWS Command Line Interface (CLI). By utilizing industry-standard OAuth, the new sign-in path facilitates a browser-based experience that integrates with AWS Identity and Access Management (IAM) federation, AWS IAM Identity Center, and traditional IAM user accounts.
The Model Context Protocol (MCP) is an open standard designed to enable AI models to interact with external tools and data sources. By adding OAuth support, AWS aims to simplify the secure connection between developer tools—such as Claude Code, Kiro, Codex, and Gemini—and AWS infrastructure. This integration ensures that agents operate within the established security perimeter of an organization while providing a streamlined login workflow.
Enhanced Security and Governance Capabilities
Alongside the OAuth rollout, AWS introduced several governance and security features intended to provide fine-grained control over agent access. These include new global condition keys specifically for OAuth, token introspection and revocation capabilities, and support for Dynamic Client Registration (DCR). The inclusion of DCR allows agents to register themselves as OAuth clients automatically, eliminating the need for manual provisioning of client IDs and secrets by developers.
For auditing and compliance, AWS has added new elements to AWS CloudTrail. These elements allow administrators to track and log authorization events related to MCP server access. Furthermore, a new API has been introduced to facilitate headless OAuth connectivity for applications and agents that do not have access to a graphical browser or operate in non-interactive environments. This ensures that the security model can be applied across various deployment scenarios, from local development workstations to automated CI/CD pipelines.
Authorization and Permissions Framework
It is important to note that authorizing an agent through OAuth does not grant that agent any additional AWS permissions beyond what the user already possesses. Every request initiated by the agent is still evaluated against existing IAM policies, Service Control Policies (SCPs), Resource Control Policies (RCPs), and permission boundaries. The OAuth token serves as a mechanism for delegated authorization, confirming the identity of the user and the intent to allow the agent to act on their behalf within the confines of established organizational controls.
Technical Implementation and Discovery
The interactive authorization model utilized by the AWS MCP Server follows several standard Request for Comments (RFC) specifications. When an agent connects for the first time, it performs a discovery process to locate the AWS Sign-In OAuth server endpoints. This is done by retrieving protected resource metadata (RFC 9728) and OAuth metadata (RFC 8414). Once the endpoints are identified, the agent uses RFC 7591 to register with AWS Sign-In and obtain a client ID.
After registration, the agent initiates a standard OAuth authorization code flow. For the developer, this results in a browser redirect to an AWS Sign-In page. If an active session already exists in the browser (for instance, if the developer is already logged into the AWS Management Console), that session can be reused. Upon approval, AWS Sign-In issues short-lived access tokens and refresh tokens. These tokens are managed automatically, allowing the agent to maintain connectivity without requiring constant manual re-authentication by the user.
Configuration Walkthrough
To utilize the new OAuth flow, developers must first ensure they have the necessary prerequisite permissions. AWS has provided a managed policy, AWSMCPSignInOAuthAccessPolicy, which must be attached to the IAM role or user intent on using the MCP server. This policy provides the specific permissions required for OAuth-based sign-in.
Setting up an agent generally involves three steps. First, the developer configures the agent to point toward the AWS MCP Server endpoint (e.g., https://aws-mcp.us-east-1.api.aws/mcp). Second, the developer reviews the authorization request triggered in the browser. Finally, once the connection is established, the developer can verify the status through the agent’s command interface, such as running a /mcp command to list active servers. Once connected, the agent can perform tasks like identifying active accounts or deploying resources using the user’s authenticated context.
Enjoyed this?
Get more posts like this delivered to your inbox.
