Thesis: Evaluation and Further Development of an MCP Server for Industrial Cameras (Waldkirch (near Freiburg), Waldkirch (near Freiburg), Waldkirch (near Freiburg), Waldkirch )

Summer Semester 2027 – fixed term of 3–6 months

Our Visionary family of 3D cameras connects to a host computer via GigE Vision/GenICam. Customers integrate the cameras using our SDK for C/C++ and Python. However, for many customers, getting started—from unboxing the camera to capturing the first image and finding the right parameters—is a time-consuming process.

We have developed a Model Context Protocol (MCP) server that allows AI assistants such as GitHub Copilot to interact with our cameras using natural language. This includes discovering and connecting to devices, setting parameters, capturing images and point clouds, and retrieving diagnostic information. An integrated simulator also allows customers to use the MCP server without a physical camera. We plan to release the MCP server as an open-source solution for our customers.

The key question is where this approach genuinely helps customers and where it falls short. Your work will investigate this question and provide evidence-based recommendations.

YOUR TASKS:

• Familiarize yourself with GigE Vision/GenICam, our SDK and the Model Context Protocol (MCP).
• Build a reproducible benchmark covering typical integration tasks, such as:
  • Setting up a camera for the first time
  • Troubleshooting network issues
  • Finding suitable parameters for a measurement task
  • Capturing point clouds
  • Developing a custom application
• Run the benchmark both in the simulator and with physical cameras.
• Systematically compare different approaches to working with the cameras:
  • Traditional development using the API and documentation
  • An AI assistant without additional tools
  • An AI assistant with domain-specific knowledge (a skill)
  • An AI assistant with device access via MCP
  • A combination of a skill and MCP
• Evaluate these approaches using criteria such as task success rate, time to first image, number of iterations, incorrect or hallucinated parameters, cost and the quality of generated code.
• Optionally, conduct a small user study with developers performing typical customer tasks.
• Improve the MCP server based on your findings. Potential areas include:
  • Bridging the gap between exploration and production code by exporting a session as an executable SDK program
  • Extending the approach to other GigE Vision devices based on their GenICam descriptions
• Analyze the opportunities and requirements for releasing the SDK on GitHub. Assess the benefits for customers and SICK, compare approaches taken by other manufacturers, and evaluate distribution, versioning and security aspects, such as prompt injection and access permissions.
• Document and present your findings and derive actionable recommendations.

YOUR PROFILE:

• Degree studies in computer science, technical computer science, electrical engineering or a comparable field
• Very good Python programming skills; C/C++ skills are an advantage
• Interest in large language models (LLMs) and AI agents
• Experience in designing empirical experiments is an advantage
• Basic knowledge of networking; knowledge of image processing or 3D data processing is an advantage
• Independent and structured working style
• Good command of German or English; the thesis can be written in English

Contact: Sarah Disch