Introduction
Artificial Intelligence is rapidly changing the way users interact with enterprise applications. Rather than navigating multiple interfaces or searching through complex systems, users increasingly expect to retrieve business information using natural language. For SAP landscapes, however, enabling this type of interaction presents a unique challenge.
SAP offers a rich ecosystem of APIs through the SAP Business Accelerator Hub, covering a wide range of business domains such as Finance, Procurement, Inventory Management, and Master Data. While these APIs provide extensive business capabilities, they are primarily designed for system-to-system integration. Their technical structure, OData entities, and complex payloads make them difficult to consume directly by AI assistants or business users.
Business users rarely think in terms of API names or technical endpoints. Instead, they think in terms of business tasks, such as:
What is the balance of G/L Account 400000?Show the details of Cost Centre IT100.List all open Purchase Orders.How much stock is available for Material MAT100?
This highlights the need for an abstraction layer that translates business requests into SAP API calls while presenting the results in a simple and meaningful way.
To address this challenge, the project was developed in two stages. The first stage focused on designing and prototyping the business tools using SAP Business Application Studio (BAS), where each tool was implemented as a CAP service with mock business data. This phase allowed the business logic, service contracts, and response structures to be validated independently of external SAP systems.
The second stage focused on integrating real SAP business data by replacing the mock implementations with live APIs from the SAP Business Accelerator Hub. These tools were then exposed through a Remote MCP Server deployed on SAP BTP Kyma, allowing AI assistants such as Claude Desktop to invoke business-oriented tools instead of interacting directly with low-level SAP APIs.
Designing Business-Oriented AI Tools
Before implementing the technical solution, the first step was to identify common business scenarios that users perform within SAP systems. Instead of exposing individual SAP APIs directly, the objective was to design reusable AI tools that represent meaningful business actions.
The tools were organized into three functional domains:
Finance – General Ledger, Cost Centres, Profit Centres, Budget Validation, and Open Accounting Items.Procurement – Purchase Orders, Goods Receipt/Invoice Receipt reconciliation, and Material Stock availability.Master Data – Business Partner information for customers and suppliers.
Each tool encapsulates one or more SAP APIs behind a business-oriented operation, allowing AI assistants to interact with SAP using business terminology rather than technical API names.
Figure 1. Overall Architecture of the AI Business Tool Platform
Prototyping the AI Business Tools with SAP Business Application Studio
The initial implementation focused on validating the business tools before integrating with live SAP systems. SAP Business Application Studio (BAS) was used to prototype each tool as an independent SAP Cloud Application Programming Model (CAP) service.
Each tool was designed with two main components:
Service Definition (.cds) – Defines the service contract by specifying the service name, the business action exposed by the tool, the required input parameters, and the structure of the returned data. For example, a Cost Centre tool may accept a Cost Centre ID as input and return information such as the Cost Centre name, Company Code, responsible person, and status.Service Implementation (.js) – Implements the business logic behind the service. It processes the input parameters, performs the required validations, retrieves or simulates the relevant business data, handles potential errors, and returns the response according to the structure defined in the CDS service.
During the prototyping phase, mock business data was used within the JavaScript implementation to simulate SAP responses. This approach enabled the service contracts, input validation, response structures, and business logic to be thoroughly tested without requiring connectivity to external SAP systems.
Once the prototypes were validated, the mock implementations could be seamlessly replaced with live SAP Business Accelerator Hub APIs while preserving the same business-oriented service interface.
CAP Service Structure
Component
Responsibility
CDS Layer (.cds)
Defines the service contract, including the exposed actions, input parameters, and response structure.
JavaScript Layer (.js)
Implements the business logic, processes requests, retrieves data, and returns the appropriate response.
Integrating SAP Business Accelerator Hub APIs
After validating the prototypes, the next step was to replace the mock business data with live SAP APIs available through the SAP Business Accelerator Hub.
The SAP Business Accelerator Hub provides a comprehensive catalog of SAP APIs covering multiple business domains, including Finance, Procurement, Human Resources, Supply Chain, and Master Data. Each API includes detailed documentation, OData service definitions, request examples, and a sandbox environment for testing.
Figure 2. SAP Business Accelerator Hub provides a centralized catalog of SAP APIs, documentation, and sandbox environments for validating business scenarios.
Reference: SAP Business Accelerator Hub.
For this project, the Business Accelerator Hub served as the primary source for identifying, validating, and integrating the SAP APIs behind each AI business tool. Although multiple APIs were integrated throughout the project, the Business Partner API is used in this article as a representative example to illustrate the overall integration process.
Figure 3. Business Partner API overview showing the available operations, documentation, and sandbox support.
Reference. Overview | Business Partner (A2X) | SAP Business Accelerator Hub
The integration process consisted of four main activities:
API Discovery
Identifying the SAP APIs that best matched each business scenario by exploring the SAP Business Accelerator Hub catalog.
API Validation
Testing the selected APIs using the SAP Business Accelerator Hub Sandbox to understand the available entities, query parameters, supported operations, and response structures before implementation.
Figure 4. The SAP Business Accelerator Hub sandbox was used to validate API requests, parameters, and expected responses before integrating the APIs into the AI tools.
Reference. Try Out | Business Partner (A2X) | SAP Business Accelerator Hub
Business Mapping
Mapping technical SAP APIs to business-oriented AI tools, ensuring that each tool represented a meaningful business operation rather than exposing technical endpoints directly.
Response Simplification
Extracting only the business-relevant fields from the SAP responses to produce concise, AI-friendly outputs while maintaining a consistent response structure.
Figure 5. Sample response returned by the Business Partner API, demonstrating the business data consumed by the AI tools.
Building the Remote MCP Server on SAP BTP Kyma
After integrating the SAP Business Accelerator Hub APIs, the next step was to expose the business tools through a Remote Model Context Protocol (MCP) Server.
The MCP server acts as an intermediary layer between AI assistants and SAP systems. Rather than allowing AI clients to invoke SAP APIs directly, the server exposes a collection of business-oriented tools that internally execute the required SAP API calls, process the responses, and return only the relevant business information.
For this project, the MCP server was deployed as a Node.js Function on SAP BTP Kyma Runtime, providing a lightweight and scalable environment for hosting AI-enabled services.
Figure 6. SAP BTP Kyma namespace hosting the Remote MCP Server and supporting Kubernetes resources.
The server implementation included the following components:
MCP Server Initialization – Configured the Remote MCP Server using the Model Context Protocol SDK.
Figure 7. Remote MCP Server deployed as a Node.js Function in SAP BTP Kyma.
Tool Registration – Registered each AI business tool together with its input schema and execution logic.
Figure 8. SAP BTP Kyma Function implementing the Remote MCP Server and registering the available AI tools.
Streamable HTTP Transport – Enabled communication between AI clients and the remote server using the MCP Streamable HTTP protocol.SAP API Integration – Connected the registered tools to the corresponding SAP Business Accelerator Hub APIs.Environment Configuration – Used Kubernetes Secrets and Environment Variables to securely provide API credentials to the application.
Figure 9. Kubernetes Secret storing the SAP API key securely outside the application source code.
This architecture allows AI assistants to invoke high-level business operations while the MCP server manages authentication, API communication, response transformation, and error handling behind the scenes.
Connecting the Remote MCP Server to Claude Desktop
With the Remote MCP Server deployed on SAP BTP Kyma, the final step was to make the business tools available to an AI assistant. For this project, Claude Desktop was used as the MCP client.
Claude Desktop supports the Model Context Protocol (MCP) and can connect to remote MCP servers through a Streamable HTTP endpoint. Once connected, Claude automatically discovers the tools exposed by the server together with their descriptions and input schemas.
The connection process involved three simple steps:
Registering the Remote MCP Server by providing the server’s Streamable HTTP endpoint.Granting tool permissions to allow Claude to invoke the available business tools.Verifying the connection, ensuring that Claude successfully discovered the exposed MCP tools.
Once the connection was established, the SAP business tools became available inside Claude Desktop as native AI tools. From that point onward, users could invoke business operations using natural language without interacting directly with SAP APIs or understanding their technical implementation.
Figure 10. Claude Desktop connected to the Remote MCP Server
This architecture demonstrates how SAP business capabilities can be securely exposed to AI assistants while keeping the integration logic centralized within the Remote MCP Server.
Conclusion
This project demonstrates how SAP Business Accelerator Hub APIs can be transformed into reusable AI business tools using the Model Context Protocol (MCP). By combining SAP Business Application Studio for prototyping, SAP BTP Kyma for deployment, and Claude Desktop as the AI client, it is possible to expose SAP business capabilities through a secure, business-oriented interface.
Rather than interacting directly with technical APIs, AI assistants invoke high-level business operations while the MCP server manages authentication, API communication, and response transformation. This architecture provides a scalable foundation for integrating enterprise SAP systems with modern AI assistants and can be extended to additional SAP business domains.
IntroductionArtificial Intelligence is rapidly changing the way users interact with enterprise applications. Rather than navigating multiple interfaces or searching through complex systems, users increasingly expect to retrieve business information using natural language. For SAP landscapes, however, enabling this type of interaction presents a unique challenge.SAP offers a rich ecosystem of APIs through the SAP Business Accelerator Hub, covering a wide range of business domains such as Finance, Procurement, Inventory Management, and Master Data. While these APIs provide extensive business capabilities, they are primarily designed for system-to-system integration. Their technical structure, OData entities, and complex payloads make them difficult to consume directly by AI assistants or business users.Business users rarely think in terms of API names or technical endpoints. Instead, they think in terms of business tasks, such as:What is the balance of G/L Account 400000?Show the details of Cost Centre IT100.List all open Purchase Orders.How much stock is available for Material MAT100?This highlights the need for an abstraction layer that translates business requests into SAP API calls while presenting the results in a simple and meaningful way.To address this challenge, the project was developed in two stages. The first stage focused on designing and prototyping the business tools using SAP Business Application Studio (BAS), where each tool was implemented as a CAP service with mock business data. This phase allowed the business logic, service contracts, and response structures to be validated independently of external SAP systems.The second stage focused on integrating real SAP business data by replacing the mock implementations with live APIs from the SAP Business Accelerator Hub. These tools were then exposed through a Remote MCP Server deployed on SAP BTP Kyma, allowing AI assistants such as Claude Desktop to invoke business-oriented tools instead of interacting directly with low-level SAP APIs.Designing Business-Oriented AI ToolsBefore implementing the technical solution, the first step was to identify common business scenarios that users perform within SAP systems. Instead of exposing individual SAP APIs directly, the objective was to design reusable AI tools that represent meaningful business actions.The tools were organized into three functional domains:Finance – General Ledger, Cost Centres, Profit Centres, Budget Validation, and Open Accounting Items.Procurement – Purchase Orders, Goods Receipt/Invoice Receipt reconciliation, and Material Stock availability.Master Data – Business Partner information for customers and suppliers.Each tool encapsulates one or more SAP APIs behind a business-oriented operation, allowing AI assistants to interact with SAP using business terminology rather than technical API names. Figure 1. Overall Architecture of the AI Business Tool PlatformPrototyping the AI Business Tools with SAP Business Application StudioThe initial implementation focused on validating the business tools before integrating with live SAP systems. SAP Business Application Studio (BAS) was used to prototype each tool as an independent SAP Cloud Application Programming Model (CAP) service.Each tool was designed with two main components:Service Definition (.cds) – Defines the service contract by specifying the service name, the business action exposed by the tool, the required input parameters, and the structure of the returned data. For example, a Cost Centre tool may accept a Cost Centre ID as input and return information such as the Cost Centre name, Company Code, responsible person, and status.Service Implementation (.js) – Implements the business logic behind the service. It processes the input parameters, performs the required validations, retrieves or simulates the relevant business data, handles potential errors, and returns the response according to the structure defined in the CDS service.During the prototyping phase, mock business data was used within the JavaScript implementation to simulate SAP responses. This approach enabled the service contracts, input validation, response structures, and business logic to be thoroughly tested without requiring connectivity to external SAP systems.Once the prototypes were validated, the mock implementations could be seamlessly replaced with live SAP Business Accelerator Hub APIs while preserving the same business-oriented service interface.CAP Service StructureComponentResponsibilityCDS Layer (.cds)Defines the service contract, including the exposed actions, input parameters, and response structure.JavaScript Layer (.js)Implements the business logic, processes requests, retrieves data, and returns the appropriate response. Integrating SAP Business Accelerator Hub APIsAfter validating the prototypes, the next step was to replace the mock business data with live SAP APIs available through the SAP Business Accelerator Hub.The SAP Business Accelerator Hub provides a comprehensive catalog of SAP APIs covering multiple business domains, including Finance, Procurement, Human Resources, Supply Chain, and Master Data. Each API includes detailed documentation, OData service definitions, request examples, and a sandbox environment for testing. Figure 2. SAP Business Accelerator Hub provides a centralized catalog of SAP APIs, documentation, and sandbox environments for validating business scenarios.Reference: SAP Business Accelerator Hub.For this project, the Business Accelerator Hub served as the primary source for identifying, validating, and integrating the SAP APIs behind each AI business tool. Although multiple APIs were integrated throughout the project, the Business Partner API is used in this article as a representative example to illustrate the overall integration process. Figure 3. Business Partner API overview showing the available operations, documentation, and sandbox support.Reference. Overview | Business Partner (A2X) | SAP Business Accelerator HubThe integration process consisted of four main activities:API DiscoveryIdentifying the SAP APIs that best matched each business scenario by exploring the SAP Business Accelerator Hub catalog.API ValidationTesting the selected APIs using the SAP Business Accelerator Hub Sandbox to understand the available entities, query parameters, supported operations, and response structures before implementation. Figure 4. The SAP Business Accelerator Hub sandbox was used to validate API requests, parameters, and expected responses before integrating the APIs into the AI tools.Reference. Try Out | Business Partner (A2X) | SAP Business Accelerator HubBusiness MappingMapping technical SAP APIs to business-oriented AI tools, ensuring that each tool represented a meaningful business operation rather than exposing technical endpoints directly.Response SimplificationExtracting only the business-relevant fields from the SAP responses to produce concise, AI-friendly outputs while maintaining a consistent response structure. Figure 5. Sample response returned by the Business Partner API, demonstrating the business data consumed by the AI tools.Building the Remote MCP Server on SAP BTP KymaAfter integrating the SAP Business Accelerator Hub APIs, the next step was to expose the business tools through a Remote Model Context Protocol (MCP) Server.The MCP server acts as an intermediary layer between AI assistants and SAP systems. Rather than allowing AI clients to invoke SAP APIs directly, the server exposes a collection of business-oriented tools that internally execute the required SAP API calls, process the responses, and return only the relevant business information.For this project, the MCP server was deployed as a Node.js Function on SAP BTP Kyma Runtime, providing a lightweight and scalable environment for hosting AI-enabled services. Figure 6. SAP BTP Kyma namespace hosting the Remote MCP Server and supporting Kubernetes resources.The server implementation included the following components:MCP Server Initialization – Configured the Remote MCP Server using the Model Context Protocol SDK. Figure 7. Remote MCP Server deployed as a Node.js Function in SAP BTP Kyma.Tool Registration – Registered each AI business tool together with its input schema and execution logic. Figure 8. SAP BTP Kyma Function implementing the Remote MCP Server and registering the available AI tools.Streamable HTTP Transport – Enabled communication between AI clients and the remote server using the MCP Streamable HTTP protocol.SAP API Integration – Connected the registered tools to the corresponding SAP Business Accelerator Hub APIs.Environment Configuration – Used Kubernetes Secrets and Environment Variables to securely provide API credentials to the application. Figure 9. Kubernetes Secret storing the SAP API key securely outside the application source code.This architecture allows AI assistants to invoke high-level business operations while the MCP server manages authentication, API communication, response transformation, and error handling behind the scenes.Connecting the Remote MCP Server to Claude DesktopWith the Remote MCP Server deployed on SAP BTP Kyma, the final step was to make the business tools available to an AI assistant. For this project, Claude Desktop was used as the MCP client.Claude Desktop supports the Model Context Protocol (MCP) and can connect to remote MCP servers through a Streamable HTTP endpoint. Once connected, Claude automatically discovers the tools exposed by the server together with their descriptions and input schemas.The connection process involved three simple steps:Registering the Remote MCP Server by providing the server’s Streamable HTTP endpoint.Granting tool permissions to allow Claude to invoke the available business tools.Verifying the connection, ensuring that Claude successfully discovered the exposed MCP tools.Once the connection was established, the SAP business tools became available inside Claude Desktop as native AI tools. From that point onward, users could invoke business operations using natural language without interacting directly with SAP APIs or understanding their technical implementation. Figure 10. Claude Desktop connected to the Remote MCP ServerThis architecture demonstrates how SAP business capabilities can be securely exposed to AI assistants while keeping the integration logic centralized within the Remote MCP Server.ConclusionThis project demonstrates how SAP Business Accelerator Hub APIs can be transformed into reusable AI business tools using the Model Context Protocol (MCP). By combining SAP Business Application Studio for prototyping, SAP BTP Kyma for deployment, and Claude Desktop as the AI client, it is possible to expose SAP business capabilities through a secure, business-oriented interface.Rather than interacting directly with technical APIs, AI assistants invoke high-level business operations while the MCP server manages authentication, API communication, and response transformation. This architecture provides a scalable foundation for integrating enterprise SAP systems with modern AI assistants and can be extended to additional SAP business domains. Read More Technology Blog Posts by Members articles
#SAP
#SAPTechnologyblog
