- SAP Managed Tags
What is Model Context Protocol?
In today’s rapidly evolving enterprise landscape, SAP business applications are at the heart of mission-critical operations—from Order to Cash and Procure to Pay, to Financial Reporting and beyond. Yet, many organizations still face persistent challenges: siloed data, fragmented workflows, and manual processes that slow innovation and limit agility.
Model Context Protocol (MCP) represents a breakthrough in how Generative AI—including Large Language Models (LLMs)—can drive real business outcomes within SAP environments. MCP transforms AI from a passive responder into an active agent, capable of planning, orchestrating, and executing complex business tasks across diverse SAP and non-SAP systems
The Model Context Protocol (MCP) enables any functionality to be exposed as AI-accessible tools, allowing models to not only understand a request but also act to fulfil it.
Born from Anthropic’s work on the Claude Code IDE, MCP was designed to seamlessly integrate a wide range of capabilities — from multi-language programming support to CI/CD frameworks and dependency management — directly into the AI’s operational toolkit. Instead of retraining models for every new capability, MCP gives them instant access to the tools they need.
Unlike Retrieval-Augmented Generation (RAG), which focuses on providing extra context to the model, MCP goes further — it provides actionable functionality. When combined, RAG and MCP form a powerful framework: RAG informs the AI with relevant knowledge, while MCP equips it with the means to execute tasks.
This synergy enables AI to plan, select the right tools, and carry out complex operations — paving the way for true agentic applications in enterprise environments. For SAP business applications, MCP is not just an enhancement — it’s the future of how AI will work with systems to drive automation, decision-making, and innovation.
How MCP works?
The Model Context Protocol (MCP) operates on client–server architecture model that transforms Large Language Models (LLMs) from passive responders into active agents capable of executing complex business tasks. On the server side, MCP hosts a library of well-defined tools — each with clear input/output specifications — that can securely access data, run processes, and return results. These servers can run locally for tasks that don’t require cloud services, or be securely hosted to connect with databases, SAP systems, and other enterprise platforms. On the client side, the MCP client acts as the bridge between the LLM and the server: it injects the right context from the user request, discovers available tools via frameworks like LangGraph, and enables the LLM to plan, select, and invoke the appropriate tool. The results are fed back to the LLM, which then produces a polished, context-aware response for the user. By connecting to multiple MCP servers, applications can tap into diverse capabilities across domains — enabling true agentic intelligence where AI autonomously plans, executes, and delivers business outcomes.
Key capabilities of MCP:
How is it Applicable to SAP Business Applications
The Model Context Protocol (MCP) has strong applicability to SAP business applications when viewed from both a business and technical perspective.
From a business perspective, SAP applications support complex, end-to-end scenarios across multiple lines of business such as Order to Cash, Procure to Pay, Hire to Retire, Materials Planning, and Financial Reporting and many more. Each of these high-level process drills down further into granular Level 2 and Level 3 processes, where specific operational steps occur. On a client landscape, the data for these processes might exist on different SAP or non-SAP systems. This poses a problem to the end-to-end process flows as this data is siloed and difficult to access.
The MCP tools can be designed to align directly with these sub-processes existing on multiple systems, enabling agentic AI clients to interpret a user’s request, extract relevant context using the tools, plan the necessary sequence of actions, and then invoke the right tools to perform operations on SAP systems. This means that tasks which traditionally require manual navigation through multiple SAP transactions can now be automated, orchestrated, and executed intelligently by AI.
From a technical perspective, SAP ecosystem offers cloud-native capabilities via SAP BTP and Kyma, which can host and run MCP servers with well-defined tools. An MCP client could, for example, can be implemented as a lightweight Flask application with multiple API routes, each tailored for specific user requests. When a request comes in, the client extracts the context, passes it to the LLM, and the LLM determines the required actions before calling the appropriate MCP server tools. This client application can be deployed on SAP BTP Cloud Foundry and integrated with a CAP application, complete with a user interface — such as BPA forms or interactive UI apps — to give end users a seamless working environment or integrated with Joule Studio
In essence, MCP server and client mechanisms can function as microservices that integrate effortlessly with SAP applications on SAP BTP, enabling:
- Direct alignment of MCP tools with SAP Level 2 and Level 3 processes.
- Secure execution of business logic via SAP ODATA services and BTP-hosted MCP servers bringing in data from other Applications into SAP.
- Context-aware orchestration of actions by LLMs, reducing manual intervention.
By combining SAP’s robust business processes with MCP’s agentic capabilities, enterprises can move toward a future where SAP applications are not only transactional but also intelligent, autonomous, and proactive in delivering business outcomes.
Example implementation
Consider a comprehensive Vendor Onboarding scenario — one that not only creates vendors in the SAP system but also validates their details via multiple external APIs, retrieves information from various sources, identifies missing data, generates a targeted questionnaire, and sends it to the vendor contact for completion.
Traditionally, this process takes several days and involves multiple teams working across different systems to collect, verify, and enter vendor details into SAP. Using Large Language Models (LLMs) combined with MCP servers and tools, this workflow can be transformed into Agentic Automation — where AI autonomously plans and executes the onboarding steps based on the specific vendor’s context. The automation adapts dynamically to each case, using available information and determining what needs to be collected, drastically reducing onboarding time while leveraging existing tools in the customer’s landscape.
In the recent SAP–AWS Hack2Build event, IBM implemented an agentic Sustainability skills using MCP server–client architecture using SAP BTP and AWS joint reference architecture as follows:
- API Availability: The necessary Sustainability Tracing APIs were already built and exposed via AWS API Gateway.
- MCP Server Setup: An MCP server was created to expose these AWS APIs, making them directly callable by the LLM. Without MCP, the APIs could only be used by other applications or services — not directly by AI agents.
- Hosting on SAP BTP: The MCP server was deployed and run on SAP BTP Cloud Foundry, ensuring secure and scalable access.
- MCP Client Application: A lightweight HTTP-routes-based MCP client was built to handle natural language queries from end users.
- The client extracts context and parameters from the user request using the LLM.
- It calls the LLM with the relevant tools and parameters so the AI can decide when to use a tool — and when not to, in case of a generic question.
- This client was also hosted on SAP BTP Cloud Foundry and connected to the MCP server to list and use tools dynamically based on the request.
- CAP Application Integration: A CAP (Cloud Application Programming) application was used as the frontend/UI layer, sending the user’s request to the MCP client and displaying results back to the user.
By simply using the MCP server and client framework hosted on SAP BTP Cloud Foundry, we built an end-to-end agentic application. This approach not only integrates external APIs directly into the SAP landscape but also provides a scalable pattern to make existing SAP applications agentic — enabling them to autonomously plan, execute, and adapt workflows.
Business Value and Benefits
Effective Agentic Automation is becoming one of the most impactful approaches to enterprise process automation, offering strategic and operational advantages across SAP landscapes. Traditionally, automation initiatives have focused on the “happy path” — the ideal scenario where processes execute exactly as planned. However, large organizations experience happy paths only about 40% of the time. The remaining majority are exceptions caused by regional variations, departmental workflows, process differences, compliance requirements, or unforeseen data gaps.
This is exactly where the Model Context Protocol (MCP) integrated with SAP business applications delivers transformative value. Rather than relying on rigid, rule-based flows, MCP enables agentic systems that adapt to context in real time. LLM-powered MCP clients can understand nuanced user requests, identify what information is available, determine what is missing, plan next steps dynamically, and call the right tools to complete the process — whether the scenario follows the happy path or not.
Applied to SAP, this means:
- End-to-end process automation for complex scenarios like Vendor Onboarding, Order to Cash, or Procure to Pay, without manual form-filling or multiple transaction hops, and integrating external tools within SAP workflows.
- Adaptive workflows that can continue operating despite ambiguity, incomplete data, or process deviations.
- Human oversight with reduced workload, where operators monitor progress and guide agents in exceptional cases, rather than executing every step themselves.
- Integration of external APIs and internal services, bringing together SAP ODATA services, SAP BTP microservices, and third-party tools under a single agentic control layer.
By hosting MCP servers and clients on SAP BTP Cloud Foundry, organizations can make existing SAP applications truly agentic — systems that don’t just respond to requests but can plan, decide, and act in pursuit of business goals. The result is a significant reduction in process cycle times, improved operational efficiency, and greater flexibility in handling real-world complexity.
In short, MCP brings the intelligence and adaptability needed to turn SAP into a proactive business partner — one that can automate not just the ideal scenarios, but also the messy, exception-driven realities that dominate enterprise operations.
Things to look out for
While the Model Context Protocol (MCP) offers powerful capabilities to make SAP applications agentic, there are several critical considerations that must be addressed to ensure secure, reliable, and scalable deployments:
- Security of Applications – MCP servers and clients can be hosted across different environments, and tools may need to access data from multiple sources. This distributed nature requires end-to-end security planning, including authentication, authorization, encryption, and secure API access. Role propagation across systems is essential so that permissions are consistently enforced throughout the workflow.
- Context Engineering – The quality and precision of context passed to the LLM directly determines which tools are invoked and in what sequence. Poorly defined context can lead to incorrect tool calls or inefficient workflows. Each use case may require careful context design and parameter mapping to ensure the LLM receives exactly the right inputs for optimal execution. This often requires iterative refinement during testing and rollout.
- Performance & Scalability – MCP-enabled workflows can involve multiple tool calls, external API requests, and data processing steps. In high-volume SAP environments, this can introduce latency if not optimized. It’s important to design for efficient tool execution, minimize unnecessary calls, cache frequently accessed data where appropriate, and ensure the MCP server infrastructure is scalable to handle concurrent requests without impacting business SLAs. Monitoring and load testing should be part of the deployment strategy.
Conclusion
The MCP Server and Client framework has the potential to transform SAP applications into truly agentic systems. By using tools exposed through MCP servers — and seamlessly integrating them with SAP BTP we can have a variety of tools that bring extensive functionality to Generative AI models — enterprises can unlock a new level of intelligent automation. With the ability to reason, plan, and decide the next steps in a process, AI agents can autonomously execute workflows while still operating under human oversight. This ensures that productivity and efficiency gains are not only substantial but also practical and measurable, delivering real business impact.
Why invest in MCP now?
As AI becomes central to enterprise strategy, MCP unlocks the potential for SAP applications to become truly intelligent, autonomous, and proactive partners in business success. By bridging the gap between AI reasoning and enterprise action, MCP empowers leaders to realize tangible ROI, accelerate innovation, and future-proof their SAP investments.
To learn more, do meet us at SAP TechEd, AWS Re:Invent or connect with us via LinkedIn.
Bluesky| User | Count |
|---|---|
| 11 | |
| 5 | |
| 5 | |
| 5 | |
| 4 | |
| 4 | |
| 4 | |
| 4 | |
| 4 | |
| 3 |
