Transitous MCP Server
Enables querying public transit data through the Transitous MOTIS API, including station search, journey routing, and generic API actions with caching and rate limiting.
README
Transitous MCP Server
A self-hostable Model Context Protocol server covering the complete public Transitous MOTIS API. It works over Streamable HTTP for ChatGPT, Claude, Grünerator, and other MCP clients, or locally over stdio.
Important: Transitous usage policy
Before deploying this server, read the Transitous API usage policy.
[!WARNING] Nutzungsrichtlinie: Die öffentliche Transitous-API wird ehrenamtlich und nach Best-Effort betrieben. Sie ist für freie/Open-Source-Anwendungen und gemeinnützige Aktivitäten gedacht, nicht standardmäßig für kommerzielle Nutzung. Vor vielen oder rechenintensiven Anfragen muss Transitous kontaktiert werden. Jede Anfrage braucht einen User-Agent mit App-Name, Version und Kontaktmöglichkeit. Quellenangaben und deren Lizenzen müssen sichtbar bleiben.
Transitous is a volunteer-run, best-effort service intended for free/open-source and non-profit use. You must:
- send a meaningful User-Agent containing the application name, version, and contact;
- cache results and avoid unnecessary or resource-intensive requests;
- contact Transitous before heavy, unusual, or commercial use;
- visibly link to Transitous data sources and preserve source-specific attribution, including OpenStreetMap attribution.
This server enforces a descriptive User-Agent, caches stop searches for 15 minutes and routing/generic API responses for 60 seconds, and limits uncached upstream calls to 20 per minute by default. Resource-intensive actions require an explicit policy acknowledgement. These protections do not grant permission beyond the Transitous policy or the individual data-source licences.
Transitous and its feeds may be incomplete, delayed, or incorrect. Do not rely on results as the sole source for critical travel decisions; verify important journeys with the relevant operator.
Tools
search_locations— resolve station/stop names to Transitous stop IDs.search_connections— find A-to-B journeys with legs, transfers, available realtime status, delays, tracks, wheelchair routing, and bicycle requirements.search_api_actions— search or list the complete allowlisted API catalog and its accepted query/body fields.execute_api_action— execute any operation from that catalog, including computational POST endpoints. Arbitrary URLs are not accepted.
All tools are read-only from the user's perspective. The upstream POST endpoints
perform routing computations; they do not create or mutate Transitous data.
Reservations and ticket sales are not part of the upstream API. The generic
plan action can request experimental fare data with withFares=true when the
underlying feeds provide it.
Complete API coverage
The catalog covers all 21 GET/POST operations in the pinned MOTIS 2.10.2 OpenAPI surface used by this release:
| Area | Actions |
|---|---|
| Journey routing | plan, refresh_itinerary_get, refresh_itinerary_post |
| Reachability | one_to_many_get, one_to_many_post, one_to_many_intermodal_get, one_to_many_intermodal_post, one_to_all |
| Places | geocode, reverse_geocode |
| Timetable | trip, stoptimes |
| Map | map_initial, map_trips, map_stops, map_levels, map_routes_experimental, map_route_details_experimental |
| Rentals | rentals |
| Operations/debug | health, debug_transfers |
The exact accepted parameters are returned by search_api_actions and validated
before any upstream request. Parameter semantics follow the
MOTIS OpenAPI document.
Experimental endpoints are clearly marked and may change upstream.
Large map and routing responses are limited to 250 kB by default and report when
they were truncated. maxResponseBytes can be raised to 1 MB, but narrowing the
geographic/time query is preferred. Actions marked resourceIntensive only run
with acknowledgeHeavyRequest=true after reviewing the Transitous policy.
Example generic action input for departures:
{
"actionId": "stoptimes",
"query": {
"stopId": "STOP_ID_FROM_SEARCH_LOCATIONS",
"n": 10,
"language": ["de"],
"withAlerts": true
}
}
Run locally
Requirements: Node.js 20+ and pnpm.
pnpm install
cp .env.example .env
pnpm build
TRANSITOUS_USER_AGENT='my-app/1.0.0 (mailto:me@example.com)' pnpm start
The default transport is stdio. Configure your local MCP client to run:
{
"mcpServers": {
"transitous": {
"command": "node",
"args": ["/absolute/path/to/transitous-mcp/dist/index.js"],
"env": {
"TRANSITOUS_USER_AGENT": "my-app/1.0.0 (mailto:me@example.com)"
}
}
}
}
Remote HTTP / Coolify
Deploy the repository with its Dockerfile, expose port 3000, and set:
MCP_TRANSPORT=http
PORT=3000
TRANSITOUS_USER_AGENT=your-app/1.0.0 (https://your.example/contact; mailto:you@example.com)
MCP_API_KEY=generate-a-long-random-secret
MCP_ALLOWED_ORIGINS=https://chatgpt.com,https://claude.ai
Endpoints:
- MCP:
https://your-domain.example/mcp - Health:
https://your-domain.example/health
Send the private key using either header:
Authorization: Bearer YOUR_SECRET
or:
X-API-Key: YOUR_SECRET
Domain/Origin filtering alone is not authentication. For a public internet
deployment, use HTTPS and set MCP_API_KEY. This version supports a static
Bearer token; OAuth can be added later for per-user authorization.
Configuration
| Variable | Required | Default | Purpose |
|---|---|---|---|
TRANSITOUS_USER_AGENT |
For tool calls | none | App name/version/contact required by Transitous |
MCP_TRANSPORT |
No | stdio |
Set http for remote deployment |
PORT |
No | 3000 |
HTTP port |
MCP_API_KEY |
Recommended for HTTP | none | Protects the MCP endpoint |
MCP_ALLOWED_ORIGINS |
No | any | Comma-separated browser origins |
TRANSITOUS_BASE_URL |
No | https://api.transitous.org |
Upstream base URL |
TRANSITOUS_REQUESTS_PER_MINUTE |
No | 20 |
Per-process uncached upstream safety limit |
MCP_REQUESTS_PER_MINUTE |
No | 60 |
Per-IP inbound HTTP limit |
Development
pnpm check
pnpm test
pnpm build
License
The MCP server code is MIT licensed. Transit data is provided by separate sources with their own licences and attribution requirements. The MIT licence does not apply to, replace, or override those data-source terms.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.