Maps turn conversational

Conversational AI is coming to maps near you. Explore how ‘conversational maps’ can unlock new user interactions across many use cases. We built the Mapbox Location Agent to show how conversational maps work. It is a location-aware conversational agent built using a large language model (LLM) and the Mapbox MCP (Model Context Protocol) Server. It engages a map as a primary component in its interactions with users. 

Talking to maps

As conversational AI becomes the primary interface for how users search, plan, and make decisions in the real world, maps can transform these interactions from passive exploration to well-reasoned and actionable guidance grounded in geospatial context. A conversational map can eliminate the multistep process of finding precise and grounded answers by combining the reasoning capabilities of LLMs with geospatial data, routing algorithms, and rich movement data layered on an interactive map.

The Mapbox Location Agent combines an LLM and the Mapbox MCP (Model Context Protocol) Server, and delivers responses to users through a map based visual interface. The LLM interprets the user’s query and the geographic entities while the Mapbox MCP Server uses Mapbox APIs to retrieve the geocodes, geometries, and directions. Finally, the agent invokes Mapbox’s map rendering and interaction libraries to layer the LLM’s response on a map. Together, they enable a conversational and interactive map interface that enables free-flowing, multi-turn conversations to support real-world tasks such as trip planning, commute comparisons, site selection, and field operations. 

Explore the Mapbox Location Agent demo by requesting access. To build your own, explore the Mapbox MCP Server on GitHub and review the Mapbox MCP Server documentation.

A whole new world for map builders

Map-based applications are typically optimized to handle specific, yet simple, queries like “hotels in Las Vegas.” Users of map-based applications have grown accustomed to common patterns such as using filtering menus or clicking map layers on and off to get answers to their questions.

However, when a map becomes conversational, many conventional, multistep user patterns are no longer necessary. Users can simply ask questions in natural language and receive visual, actionable answers tailored to their needs. Here is one illustrative example of how a conversational map handles a complex workflow end to end:

• Prompt: “What museums have special exhibits open during my trip that I could walk to in about 10 minutes from the main hotels close to Times Square?” 

Reasoning steps: The location agent geocodes Times Square, interprets an area near to it, identifies hotels within that area, builds out a 10-minute walk travel time polygon (or ‘isochrone’) around those hotel locations, queries museums that fall within the area, researches the current exhibits at the museums, and then returns the results plotted on a visual map – all in seconds.

This pattern extends to a wide range of real-world scenarios that previously required multiple apps and significant human reasoning:

• “What is the fastest way to deliver these orders today without running into bridges that the truck is too tall for?” The agent applies vehicle-specific routing rules, evaluates route options, accounts for constraints, and produces an optimized, safe route on the map.
• “Moving to Chicago for a new job – which neighborhoods have great schools, dog parks, low noise, strong safety records, and walkable access to shops and bakeries?” The agent synthesizes multiple search queries, builds a composite profile of neighborhood attributes, summarizes recommendations, and highlights candidate areas directly on the map.

Together, these examples show how conversational maps collapse multi-step workflows into a single prompt, delivering grounded, actionable, and spatially aware guidance through an intuitive map interface.

Conversational maps in action

The following demos illustrate how conversational AI is transforming user expectations for typical map interactions.

Best Sphere view in Vegas

Prompt: “What’s the best hotel room I can book that has a great view of the Sphere in Las Vegas?” 

In a conversational map: The agent analyzes viewpoints, hotel locations, and room rankings, then returns options in an immersive 3D map.

Multi-stop errand planner

Prompt: “Plan a route so I can stop at the post office, grocery store, and pharmacy on my way home from work today.” 

In a conversational map: The agent selects the best stops and optimizes the order in a single request.

Ordering Thai food to-go in Helsinki

Prompt: “Find Thai food nearby… place a pickup order… book a scooter.” 

In a conversational map: Multi-step errands across multiple apps happen in a simple conversational workflow.

Build with the Mapbox MCP Server 

Combining AI with Mapbox makes map and navigation experiences deliver smarter, more tailored recommendations to the particular needs of users. The Mapbox Location Agent is a demonstration of how maps are evolving to support easy exploration, planning, and decision-making in whole new ways. Request access to the Mapbox Location Agent to continue exploring. 

Ready to start building? Explore the Mapbox MCP Server and connect with the Mapbox team about collaborating on the future of conversational maps.