What was delivered
- An MCP server that gives AI assistants a standard set of SAP tools, so getting ERP data becomes a question in plain words instead of a transaction code or a report request.
- Built by PI console running on the private local LLM: the same agentic console colleagues use for their daily routines wrote the integration.
- Role-based access control: each user's role decides which SAP data and tools they can use, and anything outside that role is refused.
- One simple way to connect, so every colleague with an MCP-capable AI client can plug in and start asking.
How it works
- A colleague asks their AI assistant something that needs SAP data.
- The assistant calls the SAP MCP server, which knows who is asking.
- The server checks that person's role and runs only the SAP lookups the role allows.
- It reads the data from SAP and hands it back, and the assistant answers in plain words.
In plain words
MCP (Model Context Protocol) is a common plug for AI: build the SAP connection once as an MCP server, and any assistant that speaks MCP can use it. Role-based access control is the bouncer at the door: it lets each person see what their job needs, and nothing more.
Why it matters
SAP data used to sit behind specialist screens and report requests. Now anyone with the right role can ask for it directly, and the access rules live in the server's code instead of in people's heads. It also showed that the local LLM can build a real enterprise integration, not just answer questions.