Bringing AI Agents closer to SAP: What’s new in AWS for SAP MCP Server

AI agents are only as useful as the business data and processes they can reach. SAP customers told us they needed a secure, governed way to connect AI agents to business data in their SAP landscape. Working backwards from that need, we launched the AWS for SAP MCP Server earlier this year. Powered by Amazon Bedrock AgentCore, the MCP Server gives AI agents access to SAP business data through dynamic Open Data Protocol (OData) catalog discovery, contextual service hints, enterprise-grade security, dual-layer authentication, and comprehensive observability. Since general availability, we have continued to build on customer feedback. We are excited to introduce new capabilities that help customers authenticate seamlessly, reach more SAP data, run AI agents more efficiently, and deploy in more Regions:

• SAP SuccessFactors integration using the Custom Catalog feature, providing secure agentic access to HR data.
• Integration scope expanded to OData V4, which unlocks advanced OData features and exposes additional business data through published APIs.
• Identity propagation via token exchange, powered by On-Behalf-Of (OBO) access that delivers an efficient single sign-on experience. IdP Support extended to EntraID enabling enterprises to authenticate securely using OAuth and OIDC-based flows.
• AgentCore Runtime V2 Support brings elastic memory management, consistent cold start times, and serverless scale-to-zero with hardware-enforced session isolation.
• Expanded Regional Availability in São Paulo, Milan, and Spain helping meet customers where they are to reduce latency and improve interoperability within their Regions.

SAP SuccessFactors Integration

Consider a common scenario: HR teams field a steady stream of employee questions on payroll, benefits, time and personnel information. With the MCP Server connected securely to the hundreds of OData V2 and V4 APIs that SAP SuccessFactors provides, Amazon Quick lets HR teams get these answers in natural language instead of navigating multiple screens and reports. Time administrators can spot missed time entries or overtime above policy and resolve exceptions before payroll runs, while HR business partners can review a team’s leave balances, benefits enrollment and recent job changes ahead of a talent review or answer a manager’s payslip question. Agents can also answer questions from organization-wide reference data, such as benefits plan options, holiday calendars and leave policies, freeing HR teams to focus on complex cases. Connect the MCP Server to SAP SuccessFactors using the Custom Catalog feature. Because agents may handle HR data, it is best to scope the catalog to the services each use case needs, keep write tools disabled unless a workflow requires them, and use a dedicated, least-privilege user authorizations. Sample stack available here.

OData V4 Support

The MCP Server surfaces SAP OData V4 APIs in addition to V2, as discoverable tools for AI agents. Customers now have the flexibility to choose from thousands of published OData APIs exposed and available for integration in SAP systems. OData V4 in particular brings advanced data types, flexible query capabilities, and improved performance. Together, these help AI agents access an expanded dataset and retrieve compact answers efficiently from SAP. Detailed features of the OData V4 API can be found here as documented by SAP (https://blog.sap-press.com/whats-different-with-odata-4-for-sap). Among the most impactful features is cross-service navigation, which lets an agent follow navigation links across services, for example from a product to its inventory, in a single call instead of stitching together separate requests. Fewer round trips mean lower latency, fewer tool calls, and simpler agent logic. Payloads are also more efficient: JSON responses are more concise, and lower metadata overhead means fewer tokens per tool call, reducing cost and leaving more of the context window for reasoning. Advanced query options such as $apply, $filter, $orderby, and $expand let agents request summarized answers and richer data instead of pulling large record sets.

Security, Identity and Authentication

When an agent reaches SAP, who is making the request? Customers want an experience where they log in once and are authorized across multiple legs, so their identity carries through to SAP. They also need full traceability of the actions AI agents perform in their SAP applications. With identity propagation, each user’s identity is carried through every authentication leg to SAP. Users sign in once with their enterprise identity provider (IdP), and the MCP Server exchanges their token for one scoped to SAP using On-Behalf-Of (OBO) access, based on the RFC 8693 token exchange standard. SAP then applies that user’s own authorizations, logs the action against that user in the Security Audit Log, eliminating the need to store shared credentials. AWS for SAP MCP Server fully supports the On-Behalf-Of flow. It is powered by AgentCore Identity, which delivers the required SSO experience and carries the user’s identity across every authentication leg to SAP and back. In addition to IdPs like Amazon Cognito, the MCP Server now supports Microsoft EntraID, expanding the spectrum of IdP coverage. Customers can use EntraID as their IdP with the MCP Server to achieve SSO experience via the On-Behalf-Of flow or other OAuth 2.0 authentication patterns including Machine-to-Machine and User-Federation flows. For the full setup, including the Entra ID app registrations, MCP deployment, SAP prerequisites and user mapping, see Achieving single-sign-on agentic access to SAP with AWS for SAP MCP Server.

AgentCore Runtime V2 Support

The AWS for SAP MCP Server now runs on AgentCore Runtime V2, bringing meaningful improvements to performance and cost efficiency. Elastic memory management and consistent cold start times help agents respond faster, while serverless scale-to-zero ensures customers only pay for what is used. Hardware-enforced session isolation adds an additional layer of security, keeping each agent session fully separated at the infrastructure level.

Expanded Regional Availability

The AWS for SAP MCP Server is now supported in three additional Regions: sa-east-1 (São Paulo), eu-south-1 (Milan), and eu-south-2 (Spain). This now makes the MCP Server available in a total of 17 regions as listed in the documentation. Why this matters to customers: Latency – deploy close to the SAP system to keep OData round trips short; Interoperability and data residency – run in the Region where enterprise systems and compliance requirements already are; and Dependencies – the MCP Server needs AgentCore Runtime and Identity in the target Region, so check their availability before deployment.

Get Started

Deploy the MCP Server in minutes with AWS CloudFormation, then connect with an AI Agent such as Amazon Quick or a custom agent built with the Strands Agents SDK hosted in AgentCore. To get started, visit the AWS for SAP MCP Server page. Start building today. Read more on the AWS for SAP MCP Server Blog to learn how customers like Fortescue, PLDT and Harman are using AWS for SAP MCP Server to achieve their Agentic AI enhancements. To learn why AWS is the platform of choice and innovation for thousands of SAP customers, visit the AWS for SAP page.

About the Author