Overview
For a manufacturer, Siemens Teamcenter is where product truth lives: parts, bills of materials, revisions and the change process around them. This project makes that data answerable in plain language by AI assistants, without loosening the access rules that protect it.
Problems and solutions
1. Simple questions take an afternoon
The problem
Getting an answer out of Teamcenter usually means the Teamcenter client or custom code against its SOA services. A question that should take a minute, such as “what is in this assembly?” or “what changed on this part?”, takes an afternoon and a specialist.
The solution
A .NET integration service talks to Teamcenter through its SOA client libraries, with pooled sessions so requests stay fast without a fresh login for every call. On top of it sits an MCP server, so AI assistants can search items, walk a bill of materials and read change history in plain language.
2. AI access must not become a back door
The problem
Giving an assistant access to product data is only acceptable if it can never see more than the person asking is allowed to see.
The solution
Requests run as the person asking, so Teamcenter’s own permissions decide what comes back. The access design was reviewed by security before any of it touched real product data.
3. The people who run it need visibility
The problem
An integration that nobody can inspect is hard to trust and hard to support.
The solution
An admin portal gives the people who operate the service one place to see sessions, health and an audit trail.
Technical Approach
Architecture
A .NET service owns the Teamcenter connection and session pooling. An MCP layer exposes a focused set of read tools (search items, walk a bill of materials, read change records) to AI assistants.
Security
Per-user identity is carried through to Teamcenter, so authorisation is enforced by the system of record rather than re-implemented in the integration.
Quality
Test-first development: every change starts with a failing test, and the build runs the same on Windows, macOS and Linux.
Stack
.NET, Siemens Teamcenter SOA client libraries, Model Context Protocol (MCP).
What this demonstrates
This project shows how I connect AI assistants to enterprise systems of record safely: useful answers for the people who need them, with access control, auditability and testing treated as part of the product rather than an afterthought. The same approach can be applied to other PLM, ERP and engineering systems.
