Placeholder image

MCP with BTP - AI Data Enabler

| Wouter Lemaire | Tom Cenens | Marian Zeis |

MCP Joule

YouTube

Spotify

Apple

Episode #269

Introduction

In episode 269 of our SAP on Azure video podcast we talk about the Model Context Protocol being used on the SAP Business Technology Platform.

A few months ago we had my colleauge Alice on the show who introduced a tool that would allow you to convert an OData service in an MCP, a Model Context Protocol, server. Since then we had many more talks and we could see a lot of projects popping up that used this idea to create new MCP servers. The benefit are obvious: OData and MCP are a match made in heaven. What is often described has a shortcoming of OData – its chattyness and too much information – is a huge benefit in MCP, where you need a good description of the various tools.

One other project that caught not only mine, but the attention of the whole SAP community is the OData MCP bridge on BTP, the SAP Business Technology Platform. To share us more about this project, the AI Data Enabler, I am happy to have Wouter, Tom and Marian with us today. Welcome to our show.

Find all the links mentioned here: https://www.saponazurepodcast.de/episode269

Reach out to us for any feedback / questions:

#Microsoft #SAP #Azure #SAPonAzure #MCP #CopilotStudio #Teams #ClaudeDesktop

Summary created by AI

  • Introduction to MCP and Its Role in AI Integration:
  • Holger, Marian, Tom, and Wouter discussed the Model Context Protocol (MCP), its origins, and its significance as an open-source standard for connecting AI applications to various data sources and tools, highlighting its adoption in the SAP community and its benefits for documentation and integration.
    • MCP Overview and Origins: Marian explained that MCP was introduced by Anthropic about a year ago as a protocol to improve documentation and connectivity for AI applications, allowing them to interact with both local and remote data sources, APIs, and code repositories.
    • Benefits of MCP for Developers: Marian emphasized MCP’s open-source nature and its initial focus on developers, enabling access to up-to-date SAP documentation and integration with various tools, which was previously limited by outdated models.
    • MCP Server Implementations: Marian described the creation of multiple MCP servers for SAP documentation, SAP notes, and integration with SAP GUI and other systems, allowing LLM clients to search and combine documentation from different sources for best practices and technical guidance.
  • OData MCP Bridge Project on SAP BTP:
  • Tom and Wouter presented the OData MCP Bridge project, detailing how it enables MCP servers to connect AI clients like Microsoft Copilot to SAP systems via the SAP Business Technology Platform (BTP), and discussed its architecture, deployment, and technical challenges.
    • Project Overview and Use Cases: Tom explained that the OData MCP Bridge allows MCP servers deployed on SAP BTP to connect AI clients to SAP systems, supporting use cases such as querying SAP data, comparing bills of material, and integrating with Microsoft Copilot Studio.
    • Agent Creation and Tool Integration: Tom demonstrated the process of creating an agent in Microsoft Copilot Studio, adding the MCP server as a tool, and configuring it with OAuth2 authentication to securely connect to SAP systems and expose services for AI-driven queries.
    • Service Discovery and Operations: Tom described how the MCP server enables discovery of SAP services, retrieval of entity metadata, and execution of CRUD operations, with support for whitelisting APIs and combining multiple MCP servers for broader integration.
    • Single Sign-On and Security: Tom and Wouter highlighted the use of Entra ID and principal propagation for single sign-on, ensuring that user roles and authorizations are respected when accessing SAP data through the MCP server, and discussed the technical solutions for secure authentication.
    • Scalability and Future Development: Wouter outlined ongoing work to improve scalability, including plans for a database-backed version of the MCP server to support hundreds of APIs, a UI for management, and a software-as-a-service offering on the SAP Store to simplify updates and maintenance for customers.
  • Integration with Multiple AI Clients and SAP Systems:
  • Holger, Wouter, and Tom discussed the flexibility of MCP servers to connect with various AI clients such as Microsoft Copilot, Teams, and SAP Joule Studio, enabling seamless integration across different SAP systems and environments, including on-premise and cloud setups.
    • Client-Agnostic Connectivity: Holger and Wouter explained that once the MCP server is established, any compatible AI client can connect to it, regardless of the underlying SAP system or deployment model, facilitating integration with tools like Copilot Studio, Teams, and Joule Studio.
    • Connecting Multiple SAP Systems: Wouter described scenarios where MCP servers can bridge data from both cloud and on-premise SAP systems, allowing organizations to leverage AI capabilities even if some systems are not yet cloud-enabled or part of the Rise program.
    • Combining Business and External Data: Marian highlighted the power of combining business data, code, and documentation from various sources through MCP and AI clients, enabling automated data analysis and decision-making without manual architectural effort.
  • Security, Authorization, and Principal Propagation:
  • Tom and Wouter addressed the importance of security in the MCP integration, explaining how principal propagation and user authorizations are enforced to ensure that end users can only access data permitted by their SAP roles, mirroring mechanisms used in SAP Fiori and Joule AI.
    • User Role Enforcement: Tom clarified that principal propagation ensures users’ SAP authorizations and roles are respected when fetching data via MCP, preventing unauthorized access and aligning with SAP’s standard security practices.
    • Technical Implementation: Wouter explained that the MCP server leverages existing SAP Fiori Launchpad authorizations and OData API exposure, passing user tokens through the cloud connector to the backend system for secure, context-aware data access.
  • Deployment, Updates, and Customer Enablement:
  • Wouter and Holger described the deployment model for the MCP server on SAP BTP, the software-as-a-service offering on the SAP Store, and how this approach simplifies updates and enables customers to quickly start using AI with their existing SAP infrastructure.
    • SAP Store Offering: Wouter stated that the MCP server is available as a managed service on the SAP Store, allowing customers to benefit from automatic updates and reduced maintenance effort.
    • Quick Start for AI Adoption: Wouter emphasized that customers can begin leveraging AI capabilities with their current SAP systems without waiting for major upgrades, using the MCP server as a bridge to integrate with AI clients.

Links

Video index

Based on and using Hugo Cookbook Theme, Font Awesome Icons and Bulma CSS. | Thanks to Recipes4AJ to get me started. | Contact me via