GUIDE5 min read · 20 October 2026
Giving AI assistants a map of your building
Why chatbots struggle with “where is…” questions, and how MCP and spatial APIs fix it.
Ask a venue chatbot where the nearest accessible toilet is and you’ll usually get a polite non-answer. Language models are good at language and bad at geometry. Without access to the building’s actual layout, they guess or refuse.
What an assistant needs
- Search over places, with categories, synonyms and floor information.
- Routing that understands floors, closures and accessibility.
- Live context: what is open, busy, booked or out of service right now.
- Clear permissions, so a visitor-facing bot can’t see staff-only data.
Where MCP fits
The Model Context Protocol (MCP) is an open standard for giving AI applications tools and data. An MCP server for a building exposes tools such as search places, get route and nearest feature. The assistant decides when to call them, and its answer is grounded in real geometry instead of a guess.
Good uses today
- Front-desk and security copilots answering location questions instantly.
- Visitor chat on venue websites that returns a route link, not a paragraph.
- Facilities assistants that can find assets, open work orders and check room status.
Things to get right
Scope what each agent can see and do, log every tool call, and keep humans in the loop for actions with consequences, such as unlocking doors or dispatching staff. An agent answers from the map’s data, and stale data gives stale answers, so the same upkeep that keeps wayfinding accurate keeps agents accurate.