Hands‑On Experience with Amap MCP: Setup, Features, and Real‑World Use Cases

MCP (Model Context Protocol), introduced by Anthropic in November 2024, is an open protocol that standardizes communication between large language models (LLMs) and external data sources or tools, offering more flexibility and convenience than traditional function calling.

Amap has adopted MCP to expose its location‑based services (LBS) through a unified interface, covering twelve core APIs—including geocoding, reverse geocoding, IP location, weather, bike, walking, driving, transit routing, distance measurement, keyword search, nearby search, and detail search. The service is built on Server‑Sent Events (SSE), enabling real‑time, one‑way data push from server to client.

To use Amap MCP, you first need an MCP client such as Cursor. The setup process involves three steps: (1) register on the Amap Open Platform and become a certified developer; (2) create an application and obtain a key in the console; (3) add the MCP service in Cursor’s settings by editing the mcp.json file with the following content:

{
  "mcpServers": {
    "amap-amap-sse": {
      "url": "https://mcp.amap.com/sse?key=YOUR_KEY"
    }
  }
}

After configuration, the MCP service appears in Cursor, and you can switch the model to agent mode to invoke the APIs. Although Amap recommends Claude‑3.7‑sonnet, the author also tested GPT‑4o with satisfactory results. For example, asking “What good food is within 1 km of Beijing University of Posts and Telecommunications?” triggers the maps_geo and maps_around_search tools.

The article presents a detailed travel‑planning scenario: the user wants a Shanghai itinerary for the May Day holiday, including attractions, hotels, routes, and weather considerations. By feeding these requirements to the LLM, Amap MCP generates a reasonable itinerary, which the user then refines to a more economical version and adjusts the output format. The model even produces a simple HTML file to display the plan.

In summary, Amap MCP offers a straightforward, cloud‑hosted solution that requires no local deployment, provides rich LBS functionality, and allows natural‑language interaction with location data. Its current drawbacks include the inability to query multi‑day weather forecasts and occasional misselection of services, but the author expects rapid improvements as the protocol matures.