travel-mcp
Provides place geocoding and nearby point-of-interest discovery using OpenStreetMap data via Nominatim and Overpass. Supports travel planning by resolving place names or addresses to coordinates and finding nearby museums, parks, restaurants, cafes, viewpoints, and attractions sorted by distance.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
travel-mcp
An MCP server for trip planning that runs on open data only – no API keys required. It gives an AI assistant four tools: look up a place, get its weather forecast, find places nearby, and put it all together into a weather-aware day plan.
flowchart LR
Client["MCP client<br/>(Claude Desktop, Inspector, ...)"] -->|"stdio, or HTTP + Bearer token"| Server["travel-mcp"]
Server --> T1[geocode_place]
Server --> T2[get_forecast]
Server --> T3[find_nearby]
Server --> T4[plan_day]
T1 --> N["Nominatim<br/>(OpenStreetMap)"]
T2 --> O["Open-Meteo"]
T3 --> P["Overpass API<br/>(OpenStreetMap)"]
T4 --> N
T4 --> O
T4 --> PTools
Tool | What it does | Source |
| Place name or address → coordinates | Nominatim |
| Daily forecast (conditions, min/max temperature, precipitation, wind) for 1–16 days | Open-Meteo |
| Named museums, parks, restaurants, cafes, viewpoints and attractions around a point, sorted by distance | Overpass |
| Geocodes a place, checks the forecast for a date and suggests outdoor stops on good days and indoor stops (museums, cafes) when rain, strong wind or extreme temperatures are expected | all of the above |
plan_day treats a day as unsuitable for outdoor plans when precipitation probability is ≥ 50 %, precipitation is ≥ 2 mm, wind is ≥ 40 km/h, or the maximum temperature is below 5 °C or above 35 °C. The reasons are returned together with the plan so the assistant can explain its suggestion.
Related MCP server: Travel Planner MCP Server
Quick start
Requires Node.js ≥ 22.18.
git clone https://github.com/zelaplagg/travel-mcp.git
cd travel-mcp
npm install
npm run buildUse with Claude Desktop (stdio)
Add this to your Claude Desktop config (claude_desktop_config.json) and restart the app:
{
"mcpServers": {
"travel": {
"command": "node",
"args": ["/absolute/path/to/travel-mcp/dist/index.js"]
}
}
}Set TRAVEL_MCP_USER_AGENT in the env block to identify yourself to Nominatim (see Data sources).
Run as an HTTP server
HTTP mode uses the MCP Streamable HTTP transport in stateless mode and requires a bearer token.
export TRAVEL_MCP_TOKEN=$(openssl rand -hex 32)
npm run start:http # listens on 127.0.0.1:3000
curl http://127.0.0.1:3000/healthThe MCP endpoint is POST /mcp with Authorization: Bearer <token>. Requests are rate limited per client address (RATE_LIMIT_PER_MINUTE, default 60). Behind a reverse proxy the client address is the proxy's address, so rate limit there as well.
Docker
docker build -t travel-mcp .
docker run --rm -p 3000:3000 -e TRAVEL_MCP_TOKEN=$(openssl rand -hex 32) travel-mcpTry it with the MCP Inspector
npx @modelcontextprotocol/inspector node dist/index.jsConfiguration
Variable | Default | Purpose |
| – | Required in HTTP mode, at least 16 characters |
|
| HTTP bind address |
|
| Per-client request limit in HTTP mode |
|
| Sent to upstream APIs; set it to your own contact or repository URL |
| public instances | Point to self-hosted instances if you need more throughput |
Design notes
Thin MCP layer, testable core.
src/server.tsonly declares tools and validates input with zod. Everything else – API clients, the day-planning logic, rate limiting, caching, auth – lives in plain modules that receive their dependencies as arguments and are tested without network access or the MCP SDK.Polite to upstream services. Nominatim calls are spaced at least 1.1 s apart and results are cached for an hour; forecasts are cached for 10 minutes. Failed requests are retried once on network errors, HTTP 429 and 5xx, never on other 4xx.
Safe by construction. Overpass queries are built only from numbers and a fixed table of tags, never from user text. Tool errors are mapped to short messages, so upstream hostnames and stack traces are not exposed to the client. Tokens are compared in constant time.
Stateless HTTP. A new server and transport are created per request, so no state is shared between clients.
Security notes
OpenStreetMap text is untrusted. Names, addresses and websites come from data anyone can edit and end up in the model's context, so they are a prompt-injection channel. Before they reach a tool result,
src/lib/sanitize.tsremoves control, zero-width, bidi-override and Unicode tag characters (a way to hide text), caps lengths and drops websites that are not plainhttp(s)URLs. This does not stop instructions written in ordinary visible words, so clients and models must still treat tool output as data.mcp-audit evalmeasures how well a model resists that; see its README.Audited with its own tooling.
mcp-audit scan -- node dist/index.js(mcp-audit) reports no findings for this server. That checks what the server advertises (descriptions, schemas, annotations), not what its code does.No user text in queries. The Overpass query is built only from numbers and a fixed category table.
HTTP mode is authenticated and rate limited (bearer token, per-IP limits), and the server calls only the three public APIs listed below.
Development
npm run typecheck # tsc, including tests
npm test # node:test, runs the TypeScript tests directly
npm run build # emits dist/
npm run dev # run src/index.ts directly (stdio)The tests run on Node's built-in test runner with TypeScript type stripping, so they need no extra tooling. External APIs are replaced by a fake fetch.
Data sources and fair use
Geocoding and places: © OpenStreetMap contributors, ODbL. Use of the public Nominatim and Overpass instances is subject to their usage policies: identify your application, keep request rates low and cache results. For heavy use, run your own instances.
Weather: Open-Meteo, whose free API is intended for non-commercial use and requires attribution (CC BY 4.0).
License
MIT
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Your personal AI travel concierge — flights, hotels, 116M+ POIs, visas, weather & more
US weather & geo for AI agents: forecasts, alerts, earthquakes, elevation, geocoding. No keys.
US weather & geo for AI agents: forecasts, alerts, earthquakes, elevation, geocoding. No keys.
Routing, geocoding, places, matrix and map tools for location-aware AI workflows.
Related MCP Servers
- -licenseNot gradedqualityNot gradedmaintenanceEnables intelligent travel planning by combining weather forecasts and route calculations for destinations. Provides personalized travel recommendations based on weather conditions and supports trip planning with persistent conversation memory.-
- FlicenseNot gradedqualityNot gradedmaintenanceEnables AI-assisted travel planning with real-time weather forecasts, place discovery, customized itinerary generation based on interests and budget, and travel distance calculations between cities.-
- FlicenseNot gradedqualityDmaintenanceEnables travel planning through natural language, providing weather forecasts, attraction search, itinerary generation, and distance calculations using free APIs.-
- AlicenseNot gradedqualityDmaintenanceProvides live weather data (current conditions, forecasts) and city geocoding through Open-Meteo API, enabling AI assistants to answer weather queries without an API key.MIT