transitlab

Query open NYC transit data from Transit Lab on your own computer. Subway run times and slow zones, bus speeds and lanes, taxi trips, Uber and Lyft trips, service alerts. It reads our public Parquet files with DuckDB, so there is nothing to sign up for and no key.

uvx transitlab tables
uvx transitlab describe subway_stop_events
uvx transitlab sql "select route_id, count(*) as stops from subway_stop_events group by 1 order by 2 desc limit 5"
uvx transitlab sql "select * from bus_lanes limit 100" --format csv -o lanes.csv
uvx transitlab download subway_segment_times --from 2026-01 --to 2026-03 -o data

Big tables are one file a month or a day. --from and --to pick which files to read (YYYY-MM-DD or YYYY-MM). They do not trim rows inside those files; add a WHERE for that. Filtering on the year and month columns in SQL skips files the same way.

Run it on our servers

With a free key from https://transitlab.nyc/api-keys, --remote sends the query to the hosted SQL API (api.transitlab.nyc) instead of reading the files on your computer. It returns at most 10,000 rows (--limit, default 1,000) and stops after 30 seconds.

export TRANSITLAB_API_KEY=tl_live_...
uvx transitlab sql --remote "select count(*) from taxi_yellow_trips where year = 2026 and month = 8"

For AI agents

claude mcp add transitlab -- uvx transitlab mcp

transitlab mcp --config claude-desktop and --config cursor print the settings for those apps. The server has four tools: list_tables, describe_table, run_sql (read-only, 1,000 rows, 60 seconds) and example_queries.

Docs: https://docs.transitlab.nyc/start/cli-and-mcp. Data: CC BY 4.0. Credit transitlab.nyc and the original source (MTA, NYC DOT or NYC TLC).