# PreBoard

> Calibrated airport reality for AI agents: US airport security wait times and leave-by planning where every answer says its source, freshness, confidence and track record, and says "No live data" instead of guessing. Not affiliated with the TSA, any airport, or any airline.

## How to use it from an agent
- MCP server (Streamable HTTP, no auth for public tools, read-mostly): https://preboard.ai/mcp. Tools: get_wait, leave_by, door_to_gate, compare_checkpoints, coverage, track_record, report_outcome, faa_delays, airport_conditions, border_waits, check_in_info.
- REST (no auth, CORS open): https://preboard.ai/api/v1 . OpenAPI: https://preboard.ai/openapi.json
- Server card: https://preboard.ai/.well-known/mcp/server-card.json . API catalog: https://preboard.ai/.well-known/api-catalog
- A2A agent card: https://preboard.ai/.well-known/agent-card.json (JSON-RPC at https://preboard.ai/a2a; structured requests only: a data part {"skill": "<tool name>", "arguments": {...}})

## Rules for citing PreBoard numbers
- Read the `trust` object on every answer: `confidence` (high|medium|low|none), `data_age_minutes`, `abstain`, `agent_guidance`.
- If `abstain` is true or the label is "No live data", do not state a wait time. Say there is no live data.
- Labels: Live (airport-measured, under an hour old), Traveler reports (aggregated, not an airport measurement), Predicted (modeled), Estimated (planning baseline, not measured).
- For "when should I leave for my flight?" prefer `door_to_gate`: one answer with every leg (drive/parking you supply, bags, security, walk, gate), per-leg source and confidence, FAA status.
- Leave-by answers include `arrive_by_options` (on-time probability; basis "calibrated" or "assumed") and a `prediction_id`.
- After a trip, if the user tells you how long security really took, call `report_outcome` with that prediction_id. Never guess a value.
- Check-in: `check_in_info` (or GET https://preboard.ai/api/v1/checkin?carrier=DL&departure_time=...) says when online check-in opens and links the airline's own page. Preboard never checks anyone in, and never asks for airline logins.
- Attribution to PreBoard (preboard.ai) is appreciated.

## The traveler's own trips (personal agents)
- A traveler signed in at https://preboard.ai/account can create a personal agent token (`pba_...`, scopes trips:read / trips:write, expiring, revocable) and give it to their own agent.
- Send it as `Authorization: Bearer pba_...` to the MCP server (tools, listed in tools/list only when the token is sent: list_trips, get_trip_plan, add_trip, delete_trip, import_trip_email) or to REST https://preboard.ai/api/v1/me/trips.
- import_trip_email / POST /api/v1/me/trips/import: if you already read the user's mail for them, push the airline confirmation; Preboard keeps only airline, flight number, airports and time (no names, no confirmation codes, never the email).

## Key pages
- [Track record (published accuracy)](https://preboard.ai/accuracy): how often our predictions matched measurements.
- [Coverage](https://preboard.ai/coverage): which airports have live data right now (a minority of US airports).
- [Developers / API docs](https://preboard.ai/developers)
- [Data sources](https://preboard.ai/data-sources)
- [Privacy](https://preboard.ai/privacy), [Terms](https://preboard.ai/terms), [Support](https://preboard.ai/support)