What if every AI assistant could efficiently answer any question about Switzerland with the right source and no guessing?
Learn more about the Challenge
Q&A Session by Swisscom Wed 16 Sep at 15.00h
👉 signup here
Details/materials on GitHub
Problem
Large language models can search the web, but they still struggle to answer questions about Switzerland reliably. Useful information is distributed across federal, cantonal, municipal, and institutional sources in several languages. It may live in a website, API, dataset, PDF, regulation, or local service calendar. Generic search often returns the wrong jurisdiction, an outdated page, a commercial summary, or information from a neighboring country that does not apply in Switzerland.
Switzerland has excellent public information, but no simple and reusable interface that lets an LLM or agent find the right Swiss source, retrieve the relevant evidence, and use it efficiently.
Objective
Build an MCP server in a GitHub repository that makes authoritative public Swiss information accessible to AI assistants as effectively as possible. The primary goal is a working server that Swisscom can access and test during the hackathon.
Teams choose their public Swiss sources, coverage, tool design, and implementation method. The result does not need to cover all of Switzerland after two days. A focused solution that provides a credible foundation for extension and maintenance is valid.
To evaluate the submissions, we will connect every MCP server to the same MCP compatible AI client and LLM test configuration, potentially using an open source client such as OpenCode. This keeps the MCP server as the main variable. Immediate integration into myAI is not required during the event, but strong solutions may be considered for later integration.
Evaluation considers five dimensions:
- Grounding quality: correctness, authoritative sources, jurisdiction, freshness, citation support, and honest handling of unsupported questions.
- Useful Swiss coverage: the breadth and practical value of the public Swiss information the server makes accessible, including the team’s declared coverage.
- Agent efficiency: tool selection quality, number of calls, response size, token use, latency, and avoidance of unnecessary live requests.
- Operability: reproducible setup, refresh and caching design, resilience, monitoring, source etiquette, and maintainability.
- Integration readiness: a coherent MCP contract, clear documentation, extensibility, and straightforward use through standard MCP clients and the Swisscom test harness.
Automated tests can cover factual grounding, citations, unsupported queries, tool calls, response size, latency, and selected freshness checks. Manual review can assess architecture, coverage, operability, and the quality of the submitted artifact.
Your solution achieves this without requiring confidential data, production user prompts, the myAI system prompt, or a myAI sandbox.
Support for Hackers
- Experts from Swisscom will attend the hackathon.
- Swisscom will offer a virtual Q&A session before the event. Details will follow.
Technical Preferences
Teams may use a public or private GitHub repository. If the repository is private, the team must grant Swisscom’s evaluation experts access to download and test the code. The repository must include clear setup instructions and document the server’s coverage and limitations. If the solution requires authentication or other access credentials, the team must provide suitable test access through a secure channel. Secrets must never be committed to the repository.
Teams are encouraged, but not required, to release their GitHub repository publicly under a license that allows others to use, extend, and maintain the MCP server. Public release may happen during or after the hackathon. An openly available Swiss grounding server could support both private and enterprise developers building agentic systems for Switzerland. It would be a valuable contribution to the AI community, strengthen Swiss digital sovereignty, and improve the quality of answers to questions about Swiss public information.
Why hack? The primary outcome is a prototype that Swisscom can test during the hackathon and potentially integrate into myAI later. It also shows us which approaches improve grounding quality and efficiency.The wider purpose is to contribute to the AI community building agentic systems with Swiss grounding. If released publicly, an MCP server can make every chatbot that integrates it smarter about Swiss questions and help Swiss users receive better answers from agentic systems using the tool. It can serve private and enterprise developers, strengthen Swiss digital sovereignty, and improve LLM quality for questions about Swiss public information.
About the Challenge Partner
Swisscom is Switzerland’s leading ICT company, providing secure and reliable connectivity, IT, and digital services. Swiss Grounding MCP comes from Swisscom’s myAI team. Through this challenge, the team wants to explore reusable ways of making AI answers about Switzerland more reliable, grounded, and useful.