Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
제작자가 쓴 설명(영어 원문)
Lviv public transport MCP: stops, timetables, routes, and live vehicle positions. No API key.
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
필요한 것
어떤 앱을 쓰세요?
위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.
웹, 데스크톱, 모바일의 Claude에서 쓸 수 있어요:
https://api.lad.lviv.ua/mcp터미널에서 이 명령어를 실행하세요
claude mcp add --transport http timetable-api-node https://api.lad.lviv.ua/mcp프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요
{
"mcpServers": {
"timetable-api-node": {
"url": "https://api.lad.lviv.ua/mcp"
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
프로젝트의 .vscode/mcp.json에 붙여 넣으세요
{
"servers": {
"timetable-api-node": {
"type": "http",
"url": "https://api.lad.lviv.ua/mcp"
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요:
https://api.lad.lviv.ua/mcp메뉴 이름은 요금제나 시기에 따라 조금 다를 수 있어요.
선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.
이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.
아직 이 서버의 도구 목록을 읽지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 6일 점검
레지스트리 이름이 GitHub 계정 vbhjckfd 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 점검하지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요
아직 점검하지 않았어요. 엔드포인트는 차례대로 돌아가며 점검해요
아직 점검하지 않았어요. 엔드포인트를 점검할 때 도구 목록을 읽어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
제작자의 클라우드에서 실행돼요. 설치 없이 인터넷으로 연결하지만, 보내는 내용은 제작자의 서버를 거쳐요.
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(1.8.11)은 그저께 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. URL로 연결되기 때문에, 원격 MCP 서버를 지원하는 앱(예: 개발자 모드의 ChatGPT)에서도 쓸 수 있어요.
같은 카테고리에는: Spec Driven Development, Remote Desktop Commander, Dechonet MCP
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
30 security tools for AI agents: phishing links, exposure maps, DNS, SSL, PQC, history, pivots.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
Express-based API for Lviv transport timetable data with a read-only MCP endpoint.
.nvmrc)nvm use
npm install cp .env.dist .env make import # first run, and whenever the GTFS feed changes make start
make import downloads the static GTFS feed (gtfs-import-config.json) and writes the two stores the server reads: database/Timetable (LokiJS: stops, routes, shapes, first-stop schedules) and database/gtfs (SQLite: trips and calendars). Both are gitignored, so a fresh clone has no data until it runs. make import-slim re-imports only the light tables into SQLite and leaves database/Timetable alone.
All variables are optional; see .env.dist.
| Variable | Default | Purpose |
|---|---|---|
PORT | 8080 | HTTP port. |
VEHICLES_LOCATION_URL | https://track.ua-gis.com/gtfs/lviv/vehicle_position | GTFS-RT vehicle positions feed. |
TRIP_UDPDATES_URL | https://track.ua-gis.com/gtfs/lviv/trip_updates | GTFS-RT trip updates feed (the misspelling is the real name). |
MCP_PUBLIC_BASE_URL | https://api.lad.lviv.ua | Base URL used in MCP metadata (icon URL). |
NEW_RELIC_LICENSE_KEY | — | Enables New Relic APM (see Monitoring). |
NEW_RELIC_APP_NAME, NEW_RELIC_LOG_LEVEL | timetable-api-node, info | New Relic agent settings. |
PULSE_WORKER_URL, PULSE_SIGNAL_SECRET | — | Signal a Pulse worker on each /stops/:code/timetable request; both must be set. |
MCP_URL | https://api.lad.lviv.ua/mcp | Only for the timetable-mcp stdio proxy: the remote endpoint it forwards to. |
nvm use && make test
The code is TypeScript and runs straight from the .ts sources under Node's built-in type stripping — there is no build step for the server, the GTFS import or the scripts. That means only erasable syntax (no enum, namespace or parameter properties), import type for type-only imports, and relative imports that name the .ts file; tsconfig.json enforces all three.
npm run typecheck # tsc, no emit
npm run build emits just mcp/stdio.js, the published bin: Node does not strip types inside node_modules, so the npm package has to ship JavaScript. It runs on npm pack/npm publish; the output is not committed.
public/stopOverrides.js is served to browsers as-is, so it stays JavaScript and is type-checked from its JSDoc.
New Relic APM, off unless NEW_RELIC_LICENSE_KEY is set. It also carries errors and the custom metrics recorded through utils/metrics.ts.
New Relic runs as a preloaded agent, so npm start carries the flags:
node -r dotenv/config -r newrelic --import newrelic/esm-loader.mjs index.ts
dotenv/config is preloaded first so .env is populated before the agent reads its configuration. The config file is newrelic.cjs (the agent is CommonJS and this project is ESM) and holds no secrets — the key comes from the environment. /health is excluded from transactions via rules.ignore.
The account is in the EU region; its license key starts with eu01xx and the agent picks the collector from that prefix. Use the 40-character ingest license key, not an NRAK-... user API key.
Cloud Run reads the key from Secret Manager:
gcloud run services update timetable-api-node --region=europe-north1 --project=timetable-252615 --set-secrets=NEW_RELIC_LICENSE_KEY=new-relic-license-key:latest
This service exposes a public read-only MCP endpoint over Streamable HTTP. It speaks MCP revision 2026-07-28 (stateless: server/discover, a per-request _meta envelope, cacheable list results) and still serves 2025-era clients that open with initialize (2025-11-25 back to 2024-11-05).
/mcp/.well-known/mcp/server-card.json/robots.txt (non-standard comment hint)Production deployment (see cloudbuild.yaml for Cloud Run) serves REST and MCP from [api.lad.lviv.ua](https://api.lad.lviv.ua). The main site [lad.lviv.ua](https://lad.lviv.ua) is the public transport website (this repo still links there in HTML sitemap and tables for people, not for the API host). Use your own origin when running locally.
/mcp flowAn MCP client (Claude, Cursor, or the MCP SDK) talks JSON-RPC over Streamable HTTP to POST /mcp. Tool handlers reuse the same Express actions as the REST API, backed by LokiJS timetable data, GTFS SQLite (via gtfs), and live GTFS-RT feeds (for example track.ua-gis.com).
graph LR;
Client[LLM or MCP client] -->|JSON-RPC Streamable HTTP| Mcp["POST /mcp"]; Mcp --> Tools[Tool handlers]; Tools --> Actions[Express actions]; Actions --> Loki[(LokiJS)]; Actions --> Gtfs[(GTFS SQLite)]; Actions --> Rt[GTFS-RT upstream]; Loki --> Actions; Gtfs --> Actions; Rt --> Actions; Actions --> Tools; Tools --> Mcp; Mcp -->|MCP tool result| Client;
MCP Inspector (local): run npx @modelcontextprotocol/inspector, then open the UI with transport and server URL prefilled (from the inspector README):
http://localhost:6274/?transport=streamable-http&serverUrl=https%3A%2F%2Fapi.lad.lviv.ua%2Fmcp
Postman / curl: call a tool on production
POST https://api.lad.lviv.ua/mcp with Content-Type: application/json and Accept: application/json, text/event-stream — the Streamable HTTP transport rejects a request that does not accept both. The server is stateless: there is no session, so tools/call works on its own without an initialize first. The response arrives as a single SSE event: message frame carrying the JSON-RPC result. This is the 2025-era request shape; a 2026-07-28 request also sends MCP-Protocol-Version: 2026-07-28 and Mcp-Method headers, plus io.modelcontextprotocol/protocolVersion and io.modelcontextprotocol/clientCapabilities in params._meta.
curl -s https://api.lad.lviv.ua/mcp \
-H 'Content-Type: application/json' \ -H 'Accept: application/json, text/event-stream' \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_stop_realtime","arguments":{"stop_id":101}}}'
Example `tools/call` body shape:
{
"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "get_stop_realtime", "arguments": { "stop_id": 101 } } }
Successful tool responses return a natural-language text summary inside MCP content items (type: "text") — e.g. *"Stop «Площа Ринок»: 6 arrivals. Next: Т02 → «Пасічна» in 3 min."* The full structured payload is in the structuredContent field (for schema-aware clients). Each structuredContent payload follows one contract:
{
"view": "transit_realtime", "data": { "...": "tool-specific payload" }, "ui_blocks": [ { "type": "map", "data": { "center": [49.84, 24.03], "zoom": 14, "layers": { "stops": "data.stop", "vehicles": "data.arrivals" } } }, { "type": "arrival_list", "data": { "source": "data.arrivals" } } ] }
ui_blocks point into data instead of repeating it. A map block gives the centre and zoom, and layers maps stops, vehicles and polylines to a dot path in the result; every stop and vehicle there has lat/lng, vehicles also bearing. An arrival_list block names the arrivals array to render, already sorted by arrival_minutes (null = no ETA).
All tools are read-only. Stop IDs are the numeric codes on stop signs, returned as strings ("707"). Route names are the short names on vehicles ("Т30", "А1"); Latin "T30"/"A01" work too, and a bare number is read as an internal route ID. An unknown route or vehicle returns isError: true with a hint on what to pass instead.
| Tool | Arguments | Returns |
|---|---|---|
search_stops | query (string, ≥ 2 chars), limit (1–25, default 10) | Stops whose Ukrainian or English name contains every query word (inflection-tolerant: "опера" finds «Театр опери та балету»; punctuation and Latin look-alike letters are ignored, so "пл. ринок" and "Cихiв" match), with routes. |
get_stops_around_location | latitude, longitude, radius_meters (50–3000, default 1000) | Stops near a point, nearest first, with distance_meters and routes. |
get_stop_realtime | stop_id, lowfloor_only (default false) | Live arrivals: route, direction (destination), vehicle_type, arrival_minutes, vehicle position, lowfloor; scheduled: true marks timetable entries shown when no vehicle is predicted. |
find_routes_between | from_stop_id or from_lat + from_lng; to_stop_id or to_lat + to_lng; depart_at or arrive_by ("HH:MM" Lviv time or ISO 8601, default now) | Routes within a 300 m walk of each stop end (600 m of a point), best first: direct options (board_stop, alight_stop, destination, stops_count, walk at each end) and transfer_options with exactly one interchange inside a 300 m stop cluster. Each option carries schedule (timetable estimate: leave_at, per-leg board_at/alight_at, arrive_at, total_minutes) and, when planning for now, next_departures (live, from the boarding stop). Suspended routes and stops a detour skips are left out; walk_meters when the ends are a short walk apart. |
get_route_static | route_name, include_shapes (default false) | Name, type, colour, stop lists for both directions, first-stop timetable; polylines only on request. |
get_route_realtime | route_name, direction (0 or 1, optional) | Vehicles on the route with destination and next_stop (id, name, ISO arrival); headways per direction: vehicle count, live gaps_minutes front to back, scheduled_headway_minutes. |
get_nearby_vehicles | latitude, longitude, radius_meters (100–1000, default 500), route, limit (1–50, default 15), lowfloor_only (default false) | Live vehicles nearest first, with destination and distance_meters; total counts all in range. |
| get_vehicle_info | vehicle_id | One vehicle: position, route, destination, plate, upcoming stops (id, name, ISO
앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요