Resources

Changelog

API and documentation release history.

Latest release

November 20, 2026: the RailCore Developers MCP server brings these docs into Claude, Cursor, VS Code and other AI coding tools. Add it to yours. The latest change to the API itself, November 19, 2026: seat availability answers the General, Tatkal, Lower berth / senior citizen and Ladies quotas. Any other quota, Premium Tatkal included, is refused with 422 QUOTA_NOT_SUPPORTED before the seat provider is asked, and is not counted.

November 20, 2026: An MCP server for AI coding tools

  • The RailCore Developers MCP server, at https://mcp.railcore.tech/developers/mcp, gives an AI assistant these docs. It searches the guides, endpoint pages and error codes; gives an endpoint's parameters, what a request to it counts as, its scope, plans and errors; writes requests in cURL, Node.js, Python, PHP, Go and Java, signed for Growth, Scale and Enterprise keys; explains an error and how to fix it; and lists the plans.
  • It needs no sign-in and no API key. It answers from the docs and sends no request to the API, so it spends none of your requests.
  • Use RailCore in AI coding tools shows how to add it to Claude Code, Claude, Cursor, VS Code or any other MCP client.
  • The API itself did not change.

November 19, 2026: Seat availability names the quotas it answers

  • GET /v1/availability/seats, POST /v1/availability/seats/batch and GET /v1/availability/calendar answer GN (General), TQ (Tatkal), SS (Lower berth / senior citizen) and LD (Ladies). Any other quota (PT, HP, DP, HO, YU, FT) is refused with 422 QUOTA_NOT_SUPPORTED, not retryable, before the seat provider is asked, and is not counted. The message names the quota and says to check it on IRCTC.
  • Those quotas answered 422 QUOTA_NOT_AVAILABLE, "This train does not offer this quota in this class", for every train. That was not about the train: RailCore's seat source cannot check these quotas on any train, including trains that do sell Premium Tatkal.
  • Batch: an item with such a quota is not asked and not counted, and comes back in its place as a failed item with error: "QUOTA_NOT_SUPPORTED". When no item names a quota the API answers, the whole request answers 422.
  • Calendar: a Premium Tatkal calendar answered 200 with an empty days list, and was counted. It now answers 422 QUOTA_NOT_SUPPORTED, not counted.
  • QUOTA_NOT_AVAILABLE keeps its meaning for the four quotas above: the train does not offer that quota in that class for the journey.
  • Trains between stations and fares: availability_summary on GET /v1/routes/trains, and fares on GET /v1/fares/estimate and GET /v1/fares/search, are given for GN and TQ. With any other quota the routes answer lists the trains without availability, the estimate answers 404 NOT_FOUND, and the fare search lists trains without fares. Their answers did not change; the reference now says so, and the request console offers GN and TQ there.

Check your integration

If you send quota=PT (or HP, DP, HO, YU, FT) to seat availability, handle 422 QUOTA_NOT_SUPPORTED: a retry, another date or another train will not change it. Show your users that the quota is checked on IRCTC instead of reading it as the train having no seats in that quota.

November 13, 2026: Free plan: 500 requests a month, 10 a minute, 100 a day

  • The Free plan includes 500 requests a month (it was 2,000), 10 a minute (was 20) and 100 a day (was 300). It still covers every core endpoint except PNR.
  • The daily limit keeps a loop that goes wrong from spending the whole month in one afternoon. Once the day's 100 are used, requests answer 429 RATE_LIMITED until the next IST day, and the rest of the month's requests stay for the days after.
  • A Free account keeps what is left of its current month. Its next monthly refill, on the day of the month its plan started, is 500. A Free account opened from today starts with 500. The new per-minute and daily limits apply to every Free key from this release.
  • Failed requests are still not counted, and Starter, Growth, Scale and Enterprise are unchanged. Nothing about the endpoints or their answers changed in this release.

Check your integration

On Free, a key that has used the month's requests is refused with 402 CREDITS_EXHAUSTED until its next refill, and one that has used the day's with 429 RATE_LIMITED until the next IST day. X-RailCore-Credits-Remaining and X-RailCore-RateLimit-Day-Remaining show what is left. Keep to 10 requests a minute, and honor Retry-After. If you need more, Starter includes 40,000 requests a month with no daily cap.

November 13, 2026: Live speed is an estimate that moves

  • GET /v1/trains/{train_number}/live: speed_kmph is RailCore's estimate of the speed where the train was last reported, from its coaches and type, the timetable and how it has run. It is never above what the train may run at (130 km/h on LHB coaches, 110 on ICF), and lower as the train leaves a halt or nears the next. The average from the last timed station it replaces could read over 150 km/h.
  • The estimate is rounded to 5 km/h, and speed_kmph moves within 5 km/h of it every 2 seconds, the same for every caller at the same moment, as a moving train's speed does. It is still an estimate: label it as one.
  • speed_basis is estimated while the train runs (it was average_since_last_report), stationary for 0 and unavailable when the report is too old to place the train. A train running through a station it does not stop at is no longer read as standing there.

November 12, 2026: Station board window, match by station, signed webhooks

  • GET /v1/stations/{station_code}/live lists a train only when its expected or booked time at the station, arriving or leaving, falls between 2 hours ago and window_hours ahead (4 or 8; any other value reads as 8). The new window object gives from and to as India-time instants (+05:30), with hours_back and hours_ahead. A departure that crosses midnight is dated the next day. A field with no value is left out, never null: source_station_code, source_station_name, current_station_code, cancellation.to_station_code and the rest.
  • Station board: in place of a field with no value, each train carries notes, which says why. Time fields still hold only real times. The keys are arrival (Starts at New Delhi: departs only, no arrival), departure (Ends at New Delhi: arrives only, no departure), halt (Starts at New Delhi: no halt, Has not stopped here yet), platform (Platform not announced yet), last_reported (Not reported yet), position (Not started yet: at its first station, New Delhi), expected and rescheduled. notes is present only when it has at least one entry.
  • Station board: a platform the board does not name yet is filled from the railway's running status of the run, when RailCore already holds it (platform_basis: "expected"), or else the train's usual platform from the timetable ("usual", not confirmed for today); a platform on the board is "announced". On 12 November all 12 NDLS trains without a platform had one this way. A train that has started gets its last_reported_at from the same running status; one that has not started has none anywhere.
  • Station board: a train with no live forecast yet shows its booked times as the expected ones, with a delay of 0 and the new timing_basis: "scheduled"; a forecast from the live feed is timing_basis: "live". A train that has not started and has no position stands at its first station: current_station_code is that station and departed_current_station is false.
  • Station board: a train the railway moved before it starts carries rescheduled_minutes (15 or more) and notes.rescheduled, and its delay counts from the new time. 22362 at New Delhi was booked to leave at 19:10 and expected at 01:30 the next day, with a delay of 0: rescheduled_minutes: 380, Rescheduled by 6 h 20 min: now departs at 01:30 on 13 Nov. It read as on time.
  • Station board: RailCore's train catalog wins over the live feed for a train it lists. The first and last stations come from the catalog, and so do train_name and extended_train_name when the feed named other ends: the feeds reuse numbers for other local trains, and 64034 was shown as SHAKURBASTI - GHAZIABAD Local where the catalog's 64034 is the Shamli - Delhi MEMU, SMQL to DLI. A train the catalog does not list is left as the feed gives it.
  • GET /v1/routes/trains and GET /v1/journeys/search take match. city, the default and the behaviour until now, lists trains between every station of the two cities, so a search from MMCT to NDLS also lists trains from BDTS to NZM. station lists only trains between the two stations asked for, and answers 404 NO_TRAINS_FOUND, not counted, when there are none. Every answer, and every result of POST /v1/journeys/compare, carries matched_by: station or city. Each train's from_station_code, from_station_name, to_station_code and to_station_name say where it is boarded and left; they were sent but not documented.
  • Also documented for the first time: from_station_code, to_station_code, journey_date and quota on GET /v1/fares/search, and enriched on GET /v1/stations/search.
  • A 404 names what it could not find: STATION_NOT_FOUND says Station ZZZZ not found. Check the station code. An unknown train on any /v1/trains/{train_number}/… path but /live answers TRAIN_NOT_FOUND, Train 99999 not found. Check the train number., where some answered a bare NOT_FOUND. On a 400 VALIDATION_ERROR, details[].field names the parameter (date, items[2].date, match, callback_url); it said custom.
  • GET /v1/trains/{train_number}/live: date is optional. Without it the answer is for the run on its way now (one that has left its first station and is not more than 3 hours past its booked arrival; with two on the way, the one that left most recently), or else the next run to leave. journey_date says which. A finished run whose arrival at its last station was reported has confidence: "high", in the body and in X-RailCore-Confidence.
  • Live train status: the first station has no scheduled_arrival, eta, delay_arrival_minutes or actual_arrival, and the last station no scheduled_departure, etd, delay_departure_minutes or actual_departure; stop_type (ORIGIN, DESTINATION) says why, and so does the station's notes: notes.arrival at the first station, notes.departure at the last. Each was a copy of the station's other time. stations[].station_name, current_station_name, approaching_station_name and previous_station_name are written as the RailCore site writes them (New Delhi, Mumbai Central), not as upper-case booking names.
  • /schedule, /stops and /route: the same rule. The ORIGIN stop has no arrival_time and the DESTINATION stop no departure_time, and each carries notes saying why; from November 12, 2026 they copied the other time. Station names are in the site's casing. A halt's distance_km is the railway timetable's whole kilometres; a passing station's is worked out to 0.1 km.
  • Seat availability, batch and calendar: a REGRET class's confirmation_probability.text reads Waitlist full (regret), where it read Booking closed or No more booking: regret means the waitlist is full, not that booking has closed. wl_pool and wl_pool_position are now filled on batch and calendar answers too, whenever the availability text names a pool (GNWL276/WL185 gives GNWL and 276).
  • A seat enquiry for a date past the booking window (reservations open 60 days before the journey, at 8:00 AM in India; the foreign tourist quota FT is exempt) is refused with 422 BOOKING_NOT_OPEN before the seat provider is asked, and is not counted. A single-seat enquiry for a day the train does not leave the boarding station is refused with 422 TRAIN_NOT_RUNNING the same way. On the calendar, a start date past the window is 422 BOOKING_NOT_OPEN, and a window that runs past the last open date stops at it. In a batch, items past the window are not asked and not counted: each comes back in its place as a failed item with error: "BOOKING_NOT_OPEN", and when every item is past the window the whole request answers 422.
  • GET /v1/trains/{train_number}/coach-position: power cars (LPR, EOG) are listed in their places, with coach_type: "POWER_CAR" and class_name Power car or Generator car. They were left out, which left a gap in position_from_engine. New power_cars counts them and total_vehicles counts every row, locomotives included; total_coaches counts neither.
  • GET /v1/fares/unreserved: routes[].fare is a number of rupees, where it was a string such as "15.00". New from_station_code and to_station_code. from and to are still sent, but are deprecated and will be removed in a later release.
  • X-RailCore-RateLimit-Limit, -Remaining and -Reset are sent only when the plan has a per-minute limit. A plan without one (Enterprise), and a request refused before the key was read (a bad request, a missing or wrong key), no longer gets 0, 0 and 0, which read like a block. The X-RailCore-RateLimit-Day-* headers are unchanged.
  • POST /v1/jobs/route-scan and POST /v1/jobs/hopping: a callback_url must be an https URL on a public host (port 443 or 1024 and up, no user name or password, not a localhost, private, link-local, CGNAT or metadata address). It is checked when the job is submitted (400 VALIDATION_ERROR, field callback_url) and again after the host is resolved, before each send. Redirects are not followed.
  • Job callbacks are signed. The 202 answer carries webhook_secret (whsec_…), shown once. Each callback is a POST of { job_id, status, result } or { job_id, status, error } with X-RailCore-Webhook-Id, -Timestamp, -Attempt and -Signature: v1=<hex>, the HMAC-SHA256 of <timestamp>.<raw body>. After a 5xx, 408, 429, timeout or dropped connection it is sent 3 more times, after 2, 8 and 30 seconds. The job's callback_status is pending until the callback is settled, then sent or failed, and callback_attempts counts the sends. Webhooks shows how to check a signature.

Check your integration

Read window on the station board, and expect trains up to 2 hours old on it. When a field is missing, read notes for why before treating it as an error. Check timing_basis before showing an expected time as a forecast: scheduled means it is the booked time. Show rescheduled_minutes as a new departure time, not as a delay. If you need only trains between the two stations you asked for, send match=station and handle its 404. Treat a missing arrival at the first station and a missing departure at the last as "none", on live train status and on schedules. Parse routes[].fare on unreserved fares as a number, and move from from and to to from_station_code and to_station_code. Read the per-minute rate-limit headers only when they are present. If you use job callbacks, save webhook_secret and check the signature of every callback.

November 12, 2026: Push-pull trains on coach position

  • GET /v1/trains/{train_number}/coach-position: a push-pull train has a locomotive at each end, and the answer now shows it. push_pull is true, locomotives is 2, the formation ends with ENG, and the list closes with a second locomotive row. Every locomotive row carries locomotive_end: FRONT or REAR.
  • Indian Railways publishes no list of push-pull trains, so RailCore marks one on evidence and says which in push_pull_basis: train_design (an Amrit Bharat Express, built as a push-pull rake), railway_announcement (the railway announced it for that train) or rake_composition (the train's coach composition, refreshed every 15 minutes, lists a locomotive at each end).
  • Every other train answers push_pull: false and locomotives: 1 (0 for a self-propelled set). total_coaches still counts coaches only.

November 12, 2026: No nulls on schedules, coach position and live trains

  • /schedule, /stops and /route: every stop has a stop_type (ORIGIN, HALT, PASSING or DESTINATION). The first station's arrival_time is its departure time and the last station's departure_time is its arrival time; both were null.
  • The same three endpoints leave out a field the timetable has no value for. A station the train runs through has no platform_number and no typical delay, so those fields are absent there, where they were null (803 nulls on one train's full route).
  • GET /v1/trains/{train_number}/coach-position: the locomotive is the first row of every loco-hauled train, and the coaches count from 2. Trains whose coach list the railway numbers from 1, such as 12656, came back with no locomotive. A self-propelled set (Vande Bharat, MEMU) still has none.
  • Coach position also carries train_name, train_type, source_station and destination_station with their names, rake_type, total_coaches, the formation written out (ENG-SLRD-GEN-S5-…), class_summary (the rake counted by class, the locomotive first) and reversal_stations. Every row has a class_name and a coach_type.
  • GET /v1/trains/{train_number}/live: every station has an eta, an etd and both delays. The live feed times halts only, so the stations a train runs through carried null; RailCore now works their times out from the delay at the timed stations either side. The new timing_basis says which each is: reported, predicted, estimated or scheduled.
  • Live train status has no null left anywhere. A field with nothing to report is left out: actual_arrival before the train has been reported at a station, platform_number where none is named, last_reported_at before a run starts, official_notice when the railway reports nothing. The first station's arrival is its departure, and the last station's departure is its arrival.

Check your integration

On these five endpoints a field that was null is now either filled or absent. If your code tests field === null, test for a missing field as well (field == null in JavaScript covers both). On coach position, position_from_engine of a train that had no locomotive row moves up by one, and sequence is now the row number. On live train status, an eta marked estimated is RailCore's calculation, not a report.

November 12, 2026: A reason with every refusal

  • GET /v1/routes/trains and GET /v1/journeys/search answer 404 NO_TRAINS_FOUND, not counted, when no train is found between the two stations: no through train, or a station code that does not exist. They answered 200 with an empty trains list, and the request was counted. On POST /v1/journeys/compare such a route is an inline error: "NO_TRAINS_FOUND".
  • GET /v1/availability/seats: an enquiry that could never succeed says why. 422 CLASS_NOT_AVAILABLE lists the classes the train carries, 422 TRAIN_NOT_RUNNING lists the days it runs from the boarding station, and 422 BOOKING_NOT_OPEN says the date has not opened for booking yet and names the day it does (a date opens 60 days ahead, at 8:00 AM in India). They answered a retryable 503, or "does not serve the selected journey". None is counted.
  • POST /v1/availability/seats/batch: an item that names no classes is checked, and counted, in the classes its train carries. It was always checked in SL, 3A, 2A and 1A, so an item for a Shatabdi failed and counted four.
  • GET /v1/fares/unreserved: with no date the fare is for today in India. It was for tomorrow, which suburban sections do not quote, so Mumbai, Kolkata and Chennai suburban pairs failed by default. The same station as from and to answers 400 VALIDATION_ERROR.
  • GET /v1/availability/calendar: each day carries catering_charge and total_fare. fare leaves compulsory catering out, so a Rajdhani day read 2765 where the ticket costs 3165.
  • GET /v1/stations/{station_code}/live: each train carries extended_train_name, its full name, and its first and last stations are named in full (Chhatrapati Shivaji Maharaj Terminus, not C SHIVAJI MAH T).
  • /schedule, /stops and /route answer 404 TRAIN_NOT_FOUND for the numbers 00000 to 00009, which a schedule source uses for test entries.
  • Enterprise: a request counts as it does on every other plan. A seat enquiry, a live train or a route search counted as 5 requests, a station board or an availability calendar as 4, a fare as 3 and a schedule as 2; each now counts as 1. Vacant berths, hopping and route scan count as 3, 4 and 8, where they counted as 8, 10 and 25. Your balance keeps its number, so it covers more requests than it did.

Check your integration

A route search with no train is now a 404: handle NO_TRAINS_FOUND where you handled an empty list. If you call unreserved fares without date and want tomorrow's fare, send the date. Treat CLASS_NOT_AVAILABLE, TRAIN_NOT_RUNNING and BOOKING_NOT_OPEN as final: do not retry them.

November 11, 2026: Clearer errors, and 404 where nothing was found

  • A journey date that has passed in India answers 400 VALIDATION_ERROR on trains between stations, journey search, seat availability, available classes and fare search. It answered 503 UPSTREAM_UNAVAILABLE, which told clients to retry a request that could never succeed.
  • GET /v1/routes/trains and GET /v1/fares/search without date, and GET /v1/availability/classes without train_number or date, answer 400 VALIDATION_ERROR naming the parameter. They answered a 400 that did not say which one. One-letter station codes such as J (Jalna) are accepted on trains between stations.
  • GET /v1/trains/{train_number}/coach-position answers 404 TRAIN_NOT_FOUND at once for a number RailCore does not list. It waited five seconds and answered 503.
  • GET /v1/stations/{station_code}/live answers 404 STATION_NOT_FOUND for a code no station has, and GET /v1/availability/calendar answers 404 TRAIN_NOT_FOUND for an unlisted train. Both answered 200 with nothing in it, and the request was counted.
  • GET /v1/fares/estimate: class now narrows the answer to that class; it was checked and then ignored. When there is no fare for the train, journey, class and quota the request answers 404 NOT_FOUND and is not counted. It answered 200 with an empty fares list.
  • GET /v1/trains/{train_number}/live: a run that has finished carries distance_km, day, halt_minutes, latitude and longitude on every station, as a running one does; they were null. freshness.upstream_served_from_cache is left out unless the live feed says; it was always null. A 404 now says that no run was found for that date.
  • GET /v1/stations/{station_code}/live: every train's cancellation is an object with status, fully_cancelled and message, as on live train status. It was null when nothing was cancelled.
  • /schedule, /stops and /route: valid_from and valid_to are present only when the railway dated the timetable. valid_to was null on almost every train.
  • availability_summary[].last_updated_at on trains between stations, journey search and compare carries its +05:30 offset. It was the only instant in the API without one.

Check your integration

If your code reads a non-null cancellation on the station board as "cancelled", switch to cancellation.status. If it passes class to the fare estimate and expects every class back, drop class. Treat valid_from, valid_to and freshness.upstream_served_from_cache as optional, and handle 404 from coach position, the station board, the seat calendar and the fare estimate.

November 11, 2026: Fewer nulls across trains, journeys and seat availability

  • GET /v1/trains/{train_number}/live: cancellation is always an object. cancellation.status is NOT_CANCELLED, PARTIALLY_CANCELLED or CANCELLED, with the same in words in cancellation.message. It was null when nothing was cancelled.
  • speed_kmph is filled. The live feed carries no locomotive speed, so RailCore works out the average from the last station the train was timed at to its latest position; it reads 0 for a train that is standing. The new speed_basis says which. It was always null.
  • Every station has a stop_type: ORIGIN, HALT, PASSING or DESTINATION. A station the train has passed carries its actual times in eta and etd, and a passing station has its booked times in scheduled_arrival and scheduled_departure. The origin's arrival and the destination's departure stay null: they do not exist.
  • train_type, train_type_label and travel_classes come from RailCore's train catalog when the live feed leaves them out. rescheduled_minutes is 0, not null, for a start that was not moved.
  • GET /v1/trains/{train_number}/coach-position: a loco-hauled train's list starts with the locomotive (coach_name: "ENG", class_code: "LOCO", position_from_engine: 1). Every row has is_locomotive, and has_locomotive says whether the train has one.
  • /schedule, /stops and /route: classes lists the train's classes from the train catalog; it was an empty array. /route now carries catering, pantry, rake_type and each halt's platform and typical delays, as /schedule does.
  • GET /v1/routes/trains, GET /v1/journeys/search and POST /v1/journeys/compare: extended_train_name is the train's full name from RailCore's train catalog (Navjeevan SF Express, beside train_name's NAVJEEVAN SF EXP). It was always null. on_time_rating is removed.
  • GET /v1/availability/seats, POST /v1/availability/seats/batch and GET /v1/availability/calendar: confirmation_probability is filled whenever the status settles it. An available class reads 100 with the text Available, and a class that cannot be booked reads 0 with the reason. It was null unless the class was waitlisted or RAC, and it stays null only when such a class has no forecast.
  • wl_pool and wl_pool_position are present only when the railway names the booking pool. They were null on every class with no pool to name.
  • enquiry_checked_at, enquiry_completed_at and inventory_updated_at are removed from seat availability. classes[].last_updated_at still says when each figure was read, and meta.timestamp when RailCore answered. days[].alternate is removed from the calendar: it was always null.
  • fare_breakdown has a new part, rounding. The railway charges a fare's parts rounded up to the next ₹5, and a fare with any rounding came back with no breakdown at all, about one in three. Every fare the railway itemises now has one, and its parts add up to total_fare.
  • The docs' request console has a calendar for date parameters. The box still takes a typed YYYY-MM-DD.

Check your integration

If your code reads a non-null cancellation as "cancelled", switch to cancellation.status or cancellation.fully_cancelled. If it draws coaches from coach_position, skip the row where is_locomotive is true, or draw it as the engine. Stop reading on_time_rating, enquiry_checked_at, enquiry_completed_at, inventory_updated_at and days[].alternate, and treat wl_pool and wl_pool_position as optional. If you show a forecast whenever confirmation_probability is not null, check status too: an available class now reads 100 and a closed one 0. If you add up fare_breakdown yourself, include rounding.

November 10, 2026: Redesigned documentation

  • The docs have a new layout. Guides and the API reference share one sidebar, every endpoint page shows its example request and response beside the reference, and search (Ctrl K, or ⌘K on a Mac) finds any page, endpoint or error code.
  • Request samples now come in PHP, Go and Java as well as cURL, Node.js and Python. A Key header switch writes every sample with X-RailCore-Key, Authorization: Bearer or X-Api-Key.
  • The playground is now a request console. It shows the status, the request's own time, the size and what the request counted as, and it can draw a response (a live train along its route, a station board, a seat calendar) as well as show the JSON.
  • Check a signature, in the HMAC guide, builds the signing payload and the signature for any request in your browser, to compare with what your code produces.
  • Every page is also served as Markdown at its own address with .md added. Copy page, beside each title, copies it or opens it in ChatGPT, Claude, Perplexity or Gemini.
  • Every page says when it last changed.
  • Nothing about the API changed in this release.

November 7, 2026: Per-request pricing, 503 for upstream failures

  • Plans now count requests. A request to any core endpoint counts as 1: stations, trains, schedules, stops, routes, coach position, live station boards, live train status, trains between stations, journey search, seat availability, calendars, classes and fares.
  • Monthly requests: Free 2,000, Starter 40,000, Growth 1,50,000 and Scale 5,00,000 (Free became 500 on November 13, 2026). Requests and billing lists prices and every cost.
  • Paid plans have no daily cap. Free accounts keep 300 requests a day (Free became 100 a day on November 13, 2026).
  • POST /v1/availability/seats/batch counts as 1 request per distinct class per item, minimum 1 request, with 4 classes counted for an item that leaves classes out. POST /v1/journeys/compare counts as 1 request per route compared, minimum 1 request.
  • Chart status counts as 1 request, vacant berths 3 requests, berth hopping 4 requests and a route scan 8 requests. Async jobs count the same at submission; polling, listing, streaming and cancelling count as 0 requests.
  • The headers keep their names: X-RailCore-Credits-Cost and X-RailCore-Credits-Remaining now count requests. One credit is one request.
  • Upstream failures answer 503 with a Retry-After header, where they answered 502 (UPSTREAM_UNAVAILABLE) and 504 (UPSTREAM_TIMEOUT). The RailCore error body, request ID and headers now always reach you. The codes are unchanged, and the request is not counted.
  • Rate limits are per account. The per-minute budget, the Free daily budget, and the caps on /v1/charts/* and seat batches are shared by every key on the account. A new key no longer starts with limits of its own.
  • The API takes the key in X-RailCore-Key, as Authorization: Bearer, or in X-Api-Key. Two headers with different keys answer 401 AMBIGUOUS_API_KEY.
  • A signature's body hash is checked against the exact bytes you send, so indented JSON signs correctly. Before, only a compact body verified.
  • The docs playground works on every plan. Signed in, Use my account runs requests as your account with a short-lived token and no signing.

Check your integration

If you retry on HTTP 502 or 504, retry on 503 as well, after the Retry-After seconds. Better, branch on error.code: METERING_UNAVAILABLE is also a 503.

November 7, 2026: Typical delay on train schedules

  • stops[].arrival_delay_minutes and stops[].departure_delay_minutes on GET /v1/trains/{train_number}/schedule, /stops and /route were described as live delay. They are a fixed typical-delay profile: the same on every date and every request.
  • New fields typical_arrival_delay_minutes and typical_departure_delay_minutes carry the same values under names that say what they are.
  • The old names are deprecated aliases with the same values. They stay until at least December 7, 2026, 30 days after this notice.

Migration required

Read typical_arrival_delay_minutes and typical_departure_delay_minutes instead of the old names before December 7, 2026. For how late a run is now, call live train status.

November 3, 2026: Signed requests, fresh seat checks, seat calendar

  • Growth, Scale, and Enterprise keys must now sign every request with X-RailCore-Timestamp, X-RailCore-Nonce and X-RailCore-Signature. An unsigned request answers 401 SIGNATURE_INVALID and says which headers to send. Free and Starter keys are unchanged.
  • New and rotated keys on those plans come with a one-time signing secret. A key issued before November 3, 2026 has none: rotate it in the dashboard to get one. Until then its requests are refused.
  • GET /v1/availability/seats asks afresh on every request, with no RailCore cache. The upstream can still answer with a figure it read earlier, so inventory_updated_at and classes[].last_updated_at now say when the figure was read; enquiry_checked_at says when RailCore asked.
  • GET /v1/availability/calendar returned no days. It now asks each date the train runs, from date (new, default today in India) for days days, and each day says when its figure was read. Dates the train does not run on are left out.
  • A bad class or quota on the seat, calendar and fare endpoints now answers 400 VALIDATION_ERROR naming the parameter and the allowed codes, and costs nothing.

Action required on Growth and above

Sign every request. If your key was issued before November 3, 2026, rotate it to get a signing secret first.

November 3, 2026: One live train endpoint

  • GET /v1/trains/{train_number}/live now returns full live status: where the train is, how late it is, and the scheduled, expected and actual times at every station on its route, including the ones it passes without stopping. It is still a live:read operation. It counted as five credits at the time, and counts as one request since November 7, 2026.
  • date is still required. from and to are now optional: RailCore uses the train's own first and last stations, so pass them only for a train missing from the train catalog.
  • Answers can be up to 10 seconds older than the live feed: RailCore may reuse its own read of the same run, and says so in freshness.railcore_cache: reused with that read's retrieved_at. Pass max_age=0 for a fresh read. Every response now carries a Server-Timing header showing where its time went.
  • New fields: source_station_code, destination_station_code, approaching_station_code and approaching_station_name, distance_from_current_station_km, distance_to_approaching_station_km, position_anchor_station_code, started_at, start_date, scheduled_duration_minutes, rescheduled_minutes, train_type, train_type_label and travel_classes.
  • Removed fields: current_station, previous_station, route_stations, position, route, geojson and geometry_accuracy. Use current_station_code, previous_station_code, latitude, longitude and stations[] instead. speed_kmph stays but is always null.
  • In freshness, fetched_at is now retrieved_at, and railcore_cache and upstream_served_from_cache are new. stations[] and next_stop keep their fields.
  • GET /v1/trains/{train_number}/live-status and GET /v1/trains/{train_number}/running are retired. They answer 410 ENDPOINT_RETIRED with Deprecation, Sunset and a Link header naming /live, and are not counted. Their documentation pages now open the live train status page.

Migration required

Point /live-status and /running requests at /live; the /live-status query works unchanged. If you read /live before November 3, 2026, switch the removed fields to their replacements above.

October 24, 2026: Train schedules from the RailCore timetable

  • GET /v1/trains/{train_number}/schedule, /stops and /route now come from the same timetable as the RailCore site and apps, refreshed every 15 minutes.
  • Halts, times and running days follow the current timetable. Some trains gain or lose halts, and a few times move.
  • The source station's arrival_time and the destination's departure_time are always null, as documented. Some trains returned "00:00" before.
  • Station codes follow the current names, for example CSMT (was CSTM).
  • /route lists every station the train passes, with times placed between halts. Trains whose full route was unavailable before, for example 12932 and 12009, now return it.
  • Around 270 trains, mostly suburban locals, have no published running days, so running_days is [].

Check your integration

Treat null times at the ends of a route as "no time", and don't assume running_days is non-empty.

October 18, 2026: Station search ranking

  • GET /v1/stations/search returns results in a new order. A station whose name has the query as a later word now ranks with names that start with it, and busier stations win ties. For example, q=delhi now returns NDLS (New Delhi) first.
  • Ticket booking offices, out agencies, goods sheds, sidings, cabins and staff codes (names ending in OA, CB, CBO, MG, GR, SDG, CABIN and similar) now come after every passenger station. An exact code match still comes first.
  • A former name that matches the query only in part, such as Delhi Main for DLI, is no longer pinned above every other result. It is ranked with the other strong matches by how busy the station is. A former name typed in full, such as VT for CSMT, still comes first.
  • confidence values are unchanged, so results are no longer always in descending confidence order. A later-word match (0.88) can now appear above a name-prefix match (0.95).
  • station_name no longer contains repeated spaces. A few stations change, for example DE is now DELHI MG (was DELHI MG). This applies to GET /v1/stations/search and GET /v1/stations/{station_code}.
  • GET /v1/charts/vacant-berths/route-scan and route-scan jobs resolve a free-text from or to with the same rules, so they never pick a booking office or goods shed. Station codes resolve exactly as before.

Check your integration

Response shapes are unchanged. If you take the first search result as the match, or sort results by confidence, check a few common queries. For an exact station, pass its code.

October 7, 2026: Full train catalog and new train names

  • GET /v1/trains/search and GET /v1/trains/{train_number} now cover about 13,700 trains, up from about 4,100. Suburban locals are included, and the catalog is refreshed every 15 minutes.
  • Every train now carries its source station, destination station and running days. These follow the current timetable, including temporary diversions and extensions.
  • train_name is now the full published name in title case instead of an abbreviation in capitals. For example, 12951 is New Delhi Tejas Rajdhani Express (was NDLS TEJAS RAJ) and 11020 is Konark Express (was KONARK EXPRESS). display_name changes the same way.
  • Search ranking is predictable. Exact and leading train-number matches come first, then name matches, and suburban locals follow long-distance trains of equal relevance.

Check your integration

Identify trains by train_number, never by train_name. If you compare or parse the old upper-case names, switch to the number or compare case-insensitively against the new names.

September 9, 2026: One canonical station board

  • Consolidated the public station-board surface into GET /v1/stations/{station_code}/live. It remains a live:read operation. It counted as four credits at the time, and counts as one request since November 7, 2026.
  • Removed the duplicate public /departures and /trains station-board routes. Their former documentation URLs now open the live station board page.
  • Corrected lifecycle classification so a train is reported as moving only when live operational evidence supports it. Upcoming origin departures use NOT_STARTED; unobserved or overdue runs use UNKNOWN.
  • Expanded each board row with source and destination station names, normalized train type, current-station state, stop type, and cancellation details while preserving the existing response envelope.
  • Removed internal data-source identity and routing metadata from public station-board responses.

Migration required

If your integration uses a retired station-board path, change it to /v1/stations/{station_code}/live. The retired API paths now return 404.

July 30, 2026: API and developer docs launch

  • Released the production Developer API at ir.railcore.tech/v1, with operations across stations, trains, journeys, availability, fares, charts, and asynchronous jobs. The reference lists the 30 operations it has today.
  • Published complete endpoint references at railcore.tech/docs, including every request field, response schema, error, scope, cost, plan limit, and implementation note.
  • Added a direct browser playground for standard API keys with local validation, cancellation, timeout handling, response headers, latency, JSON, and server-sent event inspection.
  • Added dedicated cURL, Node.js, and Python examples to every endpoint page. Samples use environment variables and never include the key entered in the playground.
  • Published the canonical OpenAPI YAML, deep-linked prerendered routes, route-specific metadata, and complete docs sitemap coverage.
  • Added responsive navigation and an accessible light and dark documentation theme that follows the system preference and supports a saved manual choice.

July 30, 2026: Launch safeguards

  • API-key authentication includes least-privilege scopes, HMAC request signing (required on Growth, Scale and Enterprise since November 3, 2026), IP allowlists, rate limits, and request metering.
  • Shared success and error envelopes include request IDs, trace IDs, freshness metadata, stable error codes, and retry guidance.
  • Playground credentials remain in memory only and are excluded from analytics capture and session replay.

Last updated