Stadia Maps MCP Server (TypeScript)

1# Stadia Maps MCP Server (TypeScript)
2 
3This is a Model Context Protocol (MCP) server
4that provides AI assistants with access to Stadia Maps APIs for location services, geocoding, routing, and mapping capabilities.
5 
6<a href="https://glama.ai/mcp/servers/@stadiamaps/stadiamaps-mcp-server-ts">
7  <img width="380" height="200" src="https://glama.ai/mcp/servers/@stadiamaps/stadiamaps-mcp-server-ts/badge" alt="Stadia Maps Location API Server MCP server" />
8</a>
9 
10## Capabilities
11 
12- Time information anywhere on earth
13  - Current time
14  - Time zone info (ID, UTC offset, and whether a special offset like daylight saving/summer time is in effect)
15- Geocoding
16  - Address lookup
17  - Place search (POIs like restaurants, cafes, etc.)
18- Static map image generation (including markers and route overlays)
19- Routing & travel time
20- Isochrones (reachable range; both time and distance contours)
21 
22### Use cases
23 
24Here are just a few of the ways you can prompt an LLM
25to make use of the tools:
26 
27- "What time is it in Tokyo?"
28- "Make me a map showing the walking route from Depoo Turg to the Põhjala Tap Room."
29- "Is the Põhjala Tap Room open right now? Use Stadia Maps to get this information."
30- "Geocode the addresses in the attached spreadsheet."
31 
32## Quickstart
33 
34## Building
35 
36We'll use `bun` for all of our instructions,
37but you can use `npm` too.
38 
391. Run `bun install` to fetch the dependencies.
402. Then, execute the build script: `bun run build`.
41 
42The built code should be in the `build` directory now.
43 
44### Configuration
45 
46This MCP server requires a Stadia Maps API key.
47Follow the directions [here](https://docs.stadiamaps.com/authentication/#api-keys) to get your free key.
48Then, set the `API_KEY` environment variable when running the server.
49 
50## Integration Examples
51 
52API key in hand, here are a few ways you can use the MCP server.
53 
54### Claude Desktop
55 
56See [Claude Desktop Setup](claude-desktop-setup.md) for detailed instructions
57on configuring this MCP server with Claude Desktop.
58Claude Desktop is a great tool for testing out workflows.
59 
60### Warp
61 
62See [Warp Setup](warp-setup.md) for detailed instructions
63on setting up Warp with this MCP server.
64Warp is a powerful agentic environment that uses LLMs to do things on your behalf.
65Whereas you would need to upload a file to Claude,
66Warp can directly access files,
67write code, and execute programs on your computer.
68 
69Paired with the Stadia Maps MCP, it's a powerful tool for tasks like bulk geocoding,
70spatially filtering datasets, and more.
71 
72### Programmatic Integration with Anthropic SDK
73 
74For developers who want to integrate the MCP server programmatically into their TypeScript applications,
75we have an [Anthropic SDK Integration Example](examples/README.md#anthropic-sdk-demo).
76This example demonstrates how to:
77 
78- Connect to the MCP server from your code
79- Integrate with Claude's API for natural language queries
80- Enhance Claude with spatial intelligence using the Stadia Maps APIs via an MCP server
81 
82## Tips and Tricks
83 
84LLMs are far from perfect,
85and sometimes you need to put some thought into crafting prompts
86to get good results.
87 
88### When possible, be specific about which tool(s) to use
89 
90Sometimes the model needs a nudge to use the right tool.
91For example, if you ask for opening hours,
92many models will try to search the web.
93This information isn't always easily available for the model to find online.
94 
95For example, watch what a model does here:
96 
97![A model struggling to find opening hours online](opening-hours-web-search-failure.png)
98 
99You can nudge the model to use the tools in this MCP like so:
100 
101![A model using the Stadia Maps API after a different prompt](opening-hours-successful-prompt.png)
102 
103### Get creative
104 
105LLMs are capable of a lot, but they can't always do everything.
106As an example, let's look at one way to bulk geocode addresses in a spreadsheet.
107 
108First, attach an Excel sheet in Claude Desktop which has a few addresses in a column.
109 
110**Prompt:** Geocode all the addresses in this spreadsheet. Add the geocoded latitude and longitude as new columns in the sheet.
111 
112At the time of this writing, Claude cannot generate a spreadsheet attachment
113directly, but it _can_ generate an HTML page or other code which generates it for you!
114In fact, Claude Desktop may actually ask you if you want an Excel sheet download,
115but it will just go ahead and generate this for you instead.
116 
117This sort of creativity is necessary, particularly when interacting with current
118generation LLMs through the usual user interfaces.
119A lot of the power of MCPs
120is that you can interact with them through code,
121which can overcome some of these limitations to provide a delightful user experience.
122 
123### Customize it!
124 
125Just like a human, LLMs do better when given the right tools for the job.
126And just like a master craftsman, you'll probably need _some_ expert knowledge
127to know how best to combine the right tools and materials.
128 
129Additionally, models have a context limit.
130While there are advanced techniques to squeeze more out of this,
131you can only benefit by tailoring the MCP to the specific API functionality you need.
132Not to mention, it will be _much_ cheaper for you to operate with just the tools you need
133for a specific context.
134**Tool schemas are vast majority of the context cost, in a simple demo!**
135If you build the suite out to ~10 tools, you can end up with tens of thousands of input tokens.
136 
137**This project is a starting point**,
138and we've left comments throughout [`index.ts`](src/index.ts) noting our approach to building tools,
139and how you can tailor them for your use case.