Skip to main content
Glama

India Hotels MCP

CI License: Apache-2.0 Data: ODbL

Contributing · Disclaimer · Security · Changelog

An MCP server for finding, comparing and planning hotel stays across India around the places a trip actually revolves around: railway stations, airports, bus terminals, landmarks, localities and raw coordinates. It merges live prices from free meta-search sources, and adds road distance and drive time to arrival and departure points plus multi-stop stay planning.

It is standalone. Other servers (for example a train-timetable server) work with it through plain values Claude passes between them: lat/lng, Indian Railways station codes, IATA airport codes and ISO-8601 datetimes.

Tools

Full input and output schemas for every tool: docs/tools.md (generated from the server's tools/list; npm run docs:tools regenerates it and CI fails if it is out of date).

Tool

What it does

resolve_place

Name or code → coordinates (stations, airports, bus stations, landmarks, localities); or coordinates → nearest stations, airports and bus stations

search_hotels

Hotels in or near a place for given dates, merged across sources, with a section per source (its own name, rating, amenities, link and offers) and the cheapest live price in INR for the party (adults, children_ages); sort by price, rating or distance; optional drive-time limit

get_hotel_details

One hotel in full from every source, grouped by source: all offers per booking site (Booking.com, Agoda, Trip.com, MakeMyTrip, …) in INR with tax status, room, meals and links, plus each source's details (description, address, photos, amenities, review highlights, nearby places, house rules)

compare_hotels

Up to 10 hotels × up to 6 labelled places: drive or walk times, totals, ranking

travel_times

Origin × destination matrix of road km, free-flow and traffic-adjusted minutes

plan_stays

Per-stop stay planning from arrival/departure times: dates, candidates ranked by price + transfer time, leave-by times, warnings, retiring rooms

find_retiring_rooms

IRCTC railway retiring rooms at or near a station, with booking rules and portal link

get_data_sources

Status, limitations and quotas of every source; bundled dataset dates and licences

All tools are read-only; nothing books, pays or cancels.

Related MCP server: DIDA Hotel MCP

Data sources

Need

Source

Notes

Live prices

trivago official MCP

Free, no key. Cheapest advertiser per hotel (Booking.com, Agoda, MakeMyTrip, …) in INR

Live prices + availability

HotelsCasa MCP

Free, no key, true radius search. EUR, converted to INR

Live prices (unofficial)

SerpApi Google Hotels (scrapes Google; not a Google API)

Off by default; needs SERPAPI_KEY and ENABLE_UNOFFICIAL_SOURCES=true; 250 free searches/month

Live prices (unofficial)

Xotelo (derived from TripAdvisor meta-search)

Off by default; enabled only with ENABLE_UNOFFICIAL_SOURCES=true. Adds ~8 s per uncached search

FX

Frankfurter (ECB rates)

Daily reference rates

Hotel locations

OpenStreetMap snapshot (bundled)

~24k hotels/guest houses/hostels; no prices (include_unpriced)

Stations, bus stations

OpenStreetMap snapshot (bundled)

~9.2k stations with codes, ~5.3k bus stations

Airports

OurAirports (bundled)

151 Indian airports

Retiring rooms

IRCTC public station list (bundled, refreshed by hand)

356 stations; live availability needs a PNR on the IRCTC portal

Geocoding

Photon, Nominatim

Public instances, cached; Nominatim limited to 1 req/s

Routing

OSRM

FOSSGIS public instance by default, ~1 req/s, cached; self-hostable

Third-party prices are meta-search prices: they can differ at checkout and may exclude GST (includes_taxes: null means unknown). Every price carries its source, seller and fetched_at. OSM routing has no traffic data, so drive times are multiplied by METRO_TRAFFIC_MULTIPLIER (default 1.5) in the 8 largest metros and OTHER_TRAFFIC_MULTIPLIER (1.2) elsewhere; both raw and adjusted minutes are returned.

None of the free live-price sources has an SLA. Any source can be switched off with PROVIDERS_DISABLED, and a failing source never fails a search: results come back from the others with sources_failed filled in. Unofficial sources (Xotelo, and SerpApi's Google Hotels scraper) run only when ENABLE_UNOFFICIAL_SOURCES=true; check the terms that apply to you first (see DISCLAIMER.md).

git clone https://github.com/shubhamyadav8901/hotels-mcp.git && cd hotels-mcp
cp .env.example .env                # optional settings; see Configuration
docker compose up -d --build        # http://localhost:3001/mcp
curl -s localhost:3001/healthz      # {"status":"ok",...}

Or run the prebuilt image (amd64 and arm64, published for each release) without cloning:

docker run -d --name india-hotels-mcp --restart unless-stopped \
  -p 127.0.0.1:3001:3001 -e HTTP_USER_AGENT="india-hotels-mcp (+https://github.com/you)" \
  ghcr.io/shubhamyadav8901/hotels-mcp:latest

Add --env-file .env to pass optional settings. The bundled datasets (data/*.json.gz) are in the repository and the image; rebuilding them is optional (see "Rebuilding the bundled data"). Run docker compose up -d after changing .env. The server listens on port 3001 (so it can run next to other local MCP servers on 3000), bound to 127.0.0.1 only.

Connect Claude Code:

claude mcp add --transport http india-hotels http://localhost:3001/mcp

Claude Desktop and claude.ai connect to remote URLs only (Settings → Connectors); expose the server through a tunnel and add its hostname to ALLOWED_HOSTS. The HTTP server has no authentication, so put your own access control in front of any tunnel.

Inspect or test:

npx @modelcontextprotocol/inspector --cli http://localhost:3001/mcp --transport http --method tools/list

Without Docker

Requires Node 22+.

npm ci
npm run build
npm start -- --http                 # HTTP on http://localhost:3001/mcp

Or let Claude Code start it over stdio:

claude mcp add --transport stdio --env HTTP_USER_AGENT="india-hotels-mcp (+https://github.com/you)" \
  india-hotels -- node /path/to/hotels-mcp/dist/src/server.js

Self-hosted routing (optional)

The public OSRM instance is for light use. To self-host:

scripts/osrm-setup.sh northern-zone   # or "india": ~1.7 GB download, ~25 GB disk, 12+ GB Docker RAM
docker compose --profile osrm up -d
# then in .env: OSRM_URL=http://localhost:5001  OSRM_FOOT_URL=http://localhost:5002

Configuration

See .env.example. Main settings: HTTP_USER_AGENT, SERPAPI_KEY (optional), ENABLE_UNOFFICIAL_SOURCES (default false; true enables Xotelo), PROVIDERS_DISABLED, OSRM_URL, OSRM_FOOT_URL, NOMINATIM_URL, PHOTON_URL, METRO_TRAFFIC_MULTIPLIER, OTHER_TRAFFIC_MULTIPLIER, TRAIN_BUFFER_MIN (default 30), FLIGHT_BUFFER_MIN (default 120), DEFAULT_MIN_RATING_PCT (default 0 = off; default for the min_rating_pct filter, which the agent can override per request), PROVIDER_DEADLINE_MS (default 20000; each source's time limit per search), SERPAPI_MAX_PAGES (default 1) and STATE_DIR (default ~/.cache/india-hotels-mcp; holds the SerpApi monthly search count so it survives restarts; docker compose keeps it in the hotels-state volume).

HTTP mode only: PORT (default 3001), HOST (bind address, default 127.0.0.1; 0.0.0.0 in the Docker image) and ALLOWED_HOSTS (comma-separated extra hostnames accepted in the Host header, any port; localhost, 127.0.0.1 and [::1] are always accepted). Under docker compose, PORT and HOST are fixed by docker-compose.yml.

Searches are for the party given by adults (1–8) plus children_ages (up to 4 children, ages 0–17), at most 8 guests, in one room: every source is asked for that occupancy, and each quotes the offer it chooses (the room is the source's choice, so check it on the booking site). Xotelo cannot price children, so it counts them as adults, and trivago prices every child as age 6 whatever ages are given. With sort=price, Google Hotels and HotelsCasa return their cheapest results rather than their most relevant ones. min_stars (and, for Google, max_price_inr) is also applied by those sources on their own price basis, before results are merged. A town or city name (place: "Jaipur") searches the town, not its main station.

Coverage: no source returns every hotel. trivago returns ~25 hotels around a point (no radius control), so wider searches (over 3 km) query up to 13 points across the circle; HotelsCasa returns pages of 10 (up to 5 pages for wide searches); Google Hotels returns ~20 per page (SERPAPI_MAX_PAGES, default 1, one SerpApi search each); Xotelo prices only the 5 nearest of the hotels it lists. search_hotels returns a coverage entry per source (hotels returned, priced, furthest km, and what limited it), so results are not mistaken for every hotel in the radius.

Guest-rating floor: min_rating_pct (e.g. 60 = 6.0/10 = 3.0/5). Its default comes from the server setting DEFAULT_MIN_RATING_PCT (0, i.e. off, unless you set it) and shows in the tool schema, so an agent can override it per request ("at least 80%" → 80, "include unrated hotels" → 0). While it is on, hotels no source rates are left out, and the notes say how many.

Each hotel comes with a section per source, holding that source's own listing and offers:

Source

What its section adds

trivago

the cheapest booking site's price, its rating and reviews, top amenities, area, hotel page

HotelsCasa

a live-checked price with the room name, meals, refundability and availability; property type; rating

Google Hotels (SerpApi)

the lowest listed price, rating and reviews, property type, amenities, check-in and check-out times

Xotelo (TripAdvisor)

a price per booking site (Agoda, Booking.com, Trip.com…) with tax, property type, usual price range (USD)

OpenStreetMap

location and type only (no prices)

search_hotels shows up to 3 offers per source; get_hotel_details shows them all, with each source's details (description, address, phone, photos, distance to the centre, location rating, review scores, pros and cons, review topics, nearby places, important info, missing amenities, badges). For HotelsCasa it calls its free hotel-details tool; with google_prices: true it also fetches Google's page for the hotel (1 SerpApi search) for a price per booking site, address and phone. When trivago's area search leaves a hotel out, get_hotel_details asks trivago for it by name (accepted only if trivago returns the same hotel id); if the hotel still can't be found live, the prices the search returned for the same stay and party are used.

Beds: no source can filter by bed type. Where the source names the room (HotelsCasa), each price carries it in room (e.g. "Family Room", "Comfort Quadruple Room", or "Deluxe Double Room" with extra beds), so check it before booking for four adults.

Example: the cheapest rooms in Jaipur for two adults and two children:

{
  "place": "Jaipur",
  "radius_km": 6,
  "check_in": "2026-11-10",
  "check_out": "2026-11-11",
  "adults": 2,
  "children_ages": [6, 9],
  "sort": "price"
}

Development

npm run typecheck && npm test && npm run format:check   # gate before committing
npm run smoke                                          # live acceptance checks (network, a few minutes)
npm run dev                                            # watch mode, stdio

Unit tests use synthetic fixtures only and never touch the network.

Rebuilding the bundled data

Run about once a month:

HTTP_USER_AGENT="india-hotels-mcp/0.1 (you@example.com)" npm run build:data
# add -- --with-irctc to refresh the IRCTC retiring-room list (a single manual request;
# IRCTC's terms forbid automated access, so never schedule it)
# the Xotelo location-key table is rebuilt separately with: npx tsx scripts/build-xotelo-keys.ts

Limitations

  • Coverage of live prices is strong in metros and tourist towns and thin in small towns; use include_unpriced to see OpenStreetMap-only listings there.

  • Indian OTAs (MakeMyTrip, Goibibo) appear only when trivago lists them as the cheapest advertiser.

  • No public-transport routing; no live traffic.

  • IRCTC retiring-room availability and exact prices are not available without a PNR.

  • The SerpApi monthly quota counter is kept in memory and resets when the server restarts, and SerpApi's 50 searches/hour limit is not enforced locally.

Contributing

Bug reports, data corrections and new sources are welcome. Read CONTRIBUTING.md for the ground rules and checks, and the Code of Conduct. Report security issues privately as described in SECURITY.md.

Disclaimer

This is an independent project, not affiliated with trivago, HotelsCasa, TripAdvisor, Xotelo, Booking.com, Agoda, MakeMyTrip, IRCTC, Indian Railways, Google, SerpApi, OpenStreetMap or any other source. Prices are meta-search prices that may exclude GST; check with the seller before booking. Nothing is booked. See DISCLAIMER.md.

Licences

Code: Apache-2.0 (LICENSE). Bundled OpenStreetMap-derived data: ODbL 1.0, © OpenStreetMap contributors (data/LICENSE, data/NOTICE). Tool responses that use OSM data carry the attribution. Other bundled and run-time data sources are listed in NOTICE.

Available Tools

8 tools
compare_hotelsCompare hotels by travel timeA
Read-onlyIdempotent

Compares up to 10 hotels by travel time to a set of labelled places, such as tonight's arrival station, tomorrow's departure airport and attractions. Hotels are hotel_ids from search_hotels (which also supplies their cheapest price) or name + lat/lng. Places are lat+lng, station code, IATA code or place name. Returns, per hotel, the time and distance to every place, total minutes, the cheapest price from the search that found it (with its source, fetch time and dates) and a rank by total minutes. Does not fetch new prices; get_hotel_details does.

Inputs: hotels (required), places (required), mode. Returns: notes, hotels

find_retiring_roomsFind IRCTC railway retiring roomsA
Read-onlyIdempotent

Lists Indian Railways stations that have IRCTC retiring rooms (rooms and dormitories inside the station for ticketed passengers), at a station code or within a radius of a place. Returns station code, name, operator and distance, the booking rules, indicative price ranges and the IRCTC booking portal link. Live availability and exact prices are not available here: they need a Confirmed or RAC PNR on the IRCTC portal.

Inputs: lat, lng, iata, place, +2 more. Returns: notes, rules, stations, booking_url, +1 more

get_data_sourcesData sources and statusA
Read-onlyIdempotent

Lists every data source this server uses: hotel price sources, hotel location data, geocoding, routing and bundled datasets. For each it gives whether it is enabled, whether it is official or an unofficial third-party endpoint, recent success/failure, remaining quota and known limitations. Bundled datasets include their build date and licence. Does not fetch any hotel data.

Inputs: no named parameters. Returns: providers, snapshots

get_hotel_detailsOne hotel's prices and details from every sourceA
Read-onlyIdempotent

Everything the sources say about one hotel for the given dates, grouped by source: each source's own name, rating, property type, amenities and link; all its offers (booking site such as Booking.com, Agoda or MakeMyTrip; price in INR and the original currency; taxes; refundability; availability; room and meals as the source names them; link), cheapest first; and its details (description, address, phone, website, photos, distance to the centre, location rating, review scores, pros and cons, review topics, nearby places, important info such as ID rules, amenities it lacks, badges). HotelsCasa's hotel page is fetched for every call; google_prices also fetches Google's page (1 SerpApi search from a monthly quota). The hotel is a hotel_id from search_hotels or plan_stays, or a name with lat/lng. Does not book.

Inputs: check_in (required), check_out (required), lat, lng, +5 more. Returns: hotel, notes, sources, check_in, +5 more

plan_staysPlan hotel stays for an itineraryA
Read-onlyIdempotent

Plans where to stay for each stop of a trip in India. Each stay is an arrival (place + time) and the next departure (place + time); places are lat+lng, station code, IATA code or place name, so train and flight times from any source can be passed in. For every stay it works out check-in/check-out dates (early-morning arrivals book the previous night), searches hotels near the arrival point and, if elsewhere, the departure point, and ranks them by nightly price plus drive time from arrival and to departure (valued at value_of_time_inr_per_hour). Returns candidates with leave-by times that include a check-in buffer for trains or flights, IRCTC retiring-room options for 3–48 h stays at stations, and warnings for late arrivals, early departures and same-day stays. Does not book.

Inputs: stays (required), adults, min_stars, radius_km, +5 more. Returns: notes, stays

resolve_placeResolve a place in IndiaA
Read-onlyIdempotent

Turns a name or code into coordinates, or coordinates into nearby transport points. With query: matches railway station codes and names, airport IATA codes and names, bus stations, and landmarks or localities via OpenStreetMap geocoding, returning ranked matches with kind, code and lat/lng. With lat/lng: returns the nearest railway stations (within 10 km), airports with IATA codes (within 60 km) and bus stations (within 5 km), with straight-line distances. Does not search hotels.

Inputs: lat, lng, kind, limit, +1 more. Returns: notes, nearby, matches

search_hotelsSearch hotels near a placeA
Read-onlyIdempotent

Finds hotels around a place in India for given dates, with live prices from several sources (trivago, HotelsCasa, Google Hotels via SerpApi, Xotelo/TripAdvisor) merged per hotel. The place is lat+lng, an Indian Railways station code, an airport IATA code, or a place name. Each hotel has its distance, stars, best-supported guest rating and cheapest price in INR, and a section per source with that source's own name, rating, property type, amenities, link and offers (booking site, price, taxes, refundability, room and meals as the source names them). Every source is asked for one room for the party, and each quotes the offer it chooses. With max_drive_minutes, keeps only hotels within that drive time (OpenStreetMap routing with a traffic allowance) and adds drive_minutes. Results are paginated. Every offer per source is in get_hotel_details; times to other places are in compare_hotels. coverage reports what each source returned and what limited it (sources return limited pages, so the list is not exhaustive). Does not book.

Inputs: check_in (required), check_out (required), lat, lng, +14 more. Returns: notes, query, total, anchor, +5 more

travel_timesTravel times between placesA
Read-onlyIdempotent

Computes road distance and travel time from each origin to each destination in India (a matrix), by car or on foot. Each place is lat+lng, a railway station code, an airport IATA code or a place name. Returns straight-line km, road km, free-flow minutes and traffic-adjusted minutes per pair, with a warning when a point is far from any road. Does not cover public transport or flights.

Inputs: origins (required), destinations (required), mode. Returns: legs, notes

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 8 tools

    Disambiguation4/5

    Tools have distinct primary purposes (search, details, compare, plan, travel, retiring rooms, place resolution, data sources), but some overlap exists: travel_times and compare_hotels both compute travel times, and plan_stays also surfaces IRCTC retiring-room options. Descriptions explicitly clarify boundaries with 'Does not...' statements, reducing confusion.

    Naming Consistency4/5

    Names consistently use snake_case and mostly follow a verb_noun pattern (get_data_sources, resolve_place, search_hotels, get_hotel_details, compare_hotels, plan_stays, find_retiring_rooms). travel_times breaks the verb_noun pattern as a noun phrase, but the style remains uniform and readable.

    Tool Count5/5

    8 tools is well within the ideal 3–15 range for a domain covering hotel search, details, comparison, trip planning, travel times, retiring rooms, place resolution, and data source introspection. Each tool earns its place with a clear, non-redundant function.

    Completeness4/5

    The surface covers core research and planning workflows: search, details, compare, plan, travel times, retiring rooms, place resolution, and data source introspection. Booking is explicitly out of scope, and no tool provides train/flight schedules (users must supply times), but no dead ends exist for the stated purpose.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables multi-modal travel planning for India, aggregating flights, trains, buses, cabs, and personal vehicles across 117 cities with multi-leg routing and weighted scoring.
      ISC
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to search, compare, and book hotels with real-time pricing and availability, supporting multiple location types, star ratings, and price filters.
      12
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Official MCP server for MAQAMI, a hotel and flight booking platform with 3M+ hotels. Search live hotel rates and flights, look up places, airports and hotel details, then prebook and book. Remote Streamable HTTP endpoint, no API key required.
      41
      101 npm
      2
      MIT