Open source · Python · MCP

Flight tracking your
agent can act on

SkyBuddy watches routes around the clock, measures every fare against a full year of its own history, and hands your agent an auditable booking intent for the exact flight link — with a separate travel workspace for every traveller.

★ Star on GitHub See agent booking

★ — stars ⑂ — forks 30 agent tools Multi-user

Traveller

Interactive preview — sample fares, stored in your browser. The Python side keeps the real per-user history.

7booking sources per search

30agent-callable tools

∞travellers & routes in parallel

365dprice history behind every buy call

Everything it does

One toolkit for the whole trip

Search, watch, score, advise, book. Each capability is an independent Python module you can call directly, from the CLI, or through the MCP server.

🔎

Real-time flight search

Live fares through the Duffel API plus deep links for Google Flights, Avianca, Iberia, KLM, Air France, Kayak and Skyscanner. flight_scraper.py

👥

Multi-user workspaces

Every traveller gets isolated preferences, watched routes, passengers, cards, bookings and price history. users.py

📈

Price monitoring

Watch unlimited routes, keep a full price history, and compute medians and trends per route. flight_monitor.py

🔔

Smart alerts

Fires on target price, below-median deals and sharp drops. Console or email, with de-duplicated alert history. alerts.py

🧠

Recommendation engine

Every fare scored 0–100 — 40% price, 30% duration, 20% stops, 10% your preferences — with written reasons. recommendations.py

📉

365-day price baseline

Scans a year of recorded fares for the route, then turns a live price into a verdict: buy now, good, fair, wait or high. price_baseline.py

💺

Seat selection, end to end

FlightAware and SeatMaps for the verified workflow, cross-checked on SeatGuru, aeroLOPA, ExpertFlyer and Flightradar24. seat_advisor.py

📅

Flexible-date sweeps

Every outbound/return pair and both baggage modes on Google Flights, keeping all carrier combinations. google_flights_monitor.py

💳

Loyalty & passengers

Cards, balances, tiers and points estimation, plus traveller profiles ready for checkout. loyalty_cards.py · passenger_profiles.py

The part everyone asks for

Let the agent buy the ticket

SkyBuddy does not scrape a checkout behind your back. It builds a booking intent: a structured, auditable object carrying the exact flight link, the passengers, the price you agreed to, and the checks that must pass before money moves.

0
auto_book_if_dealOptional: the 365-day baseline decides the fare is buy-worthy and creates the intent for you.
1
prepare_bookingPins the chosen offer, its booking URL and price into a saved intent, inside that traveller's workspace.
2
Human confirmationNothing executes until you approve the intent by its id.
3
confirm_bookingRe-validates the price ceiling, then releases the agent playbook.
4
Agent executesOpens the link, fills passenger data, checks the seat map on long-haul, and stops at payment for the final human OK.

Safe by default: dry-run mode, a hard price ceiling, and an explicit stop before any payment step.

booking_agent.py

from agent_integration import create_agent

sky = create_agent(user="harold")   # own workspace

intent = sky.prepare_booking(
    booking_url="https://www.iberia.com/…",
    airline="Iberia",
    origin="BIO", destination="BOG",
    outbound_date="2026-12-04",
    return_date="2027-01-08",
    price=684.0, currency="EUR",
    passengers=["harold"],
    max_price=700.0,          # hard ceiling
)

# → {"intent_id": "bk_7f3a1c",
#    "status": "awaiting_confirmation"}

sky.confirm_booking("bk_7f3a1c", approved_by="harold")

# → "ready_to_execute" + a step-by-step
#   agent_playbook for the booking link

Quick start

Running in under a minute

Pure Python, one required dependency. Pick your entry point.

bash

git clone https://github.com/HaroldMate1/SkyBuddy.git
cd SkyBuddy
pip install -r requirements.txt

# search a route (config or arguments)
python scripts/flight_scraper.py BIO BOG 2026-12-04 2027-01-08 EUR

# monitor every watched route and raise alerts
python scripts/flight_monitor.py

# what does a good price look like? (365 days by default)
python scripts/price_baseline.py scan --origin BIO --destination BOG
python scripts/price_baseline.py evaluate --origin BIO --destination BOG --price 684

# create a workspace per traveller
python scripts/users.py create --user harold --display-name "Harold Mateo" \
  --email harold@example.com --currency EUR --home-airport BIO
python scripts/users.py create --user ana --display-name "Ana" --currency EUR

python scripts/users.py list
python scripts/users.py switch --user ana      # set the active traveller
python scripts/users.py show --user ana        # profile + workspace paths

# every command runs inside the active workspace, or an explicit one
python scripts/flight_monitor.py --user harold
python scripts/booking_agent.py list --user ana

# each traveller keeps their own routes, alerts, bookings and history:
#   data/users/harold/{price_baseline.csv,bookings.json,alerts.json}
#   config/users/harold/{preferences.json,passengers.json,cards.json}

from agent_integration import create_agent

sky = create_agent(user="harold")

result = sky.search_flights(
    origin="BIO", destination="BOG",
    outbound_date="2026-12-04",
    return_date="2027-01-08",
)

for rec in result["top_recommendations"]:
    print(rec["score"], rec["airline"], rec["price"], rec["booking_url"])

sky.add_route("Colombia Trip", "BIO", "BOG",
              "2026-12-04", "2027-01-08", target_price=650)
sky.check_all_routes()          # only harold's routes

# start the universal MCP server
python scripts/mcp_server.py claude

# claude_desktop_config.json / .mcp.json
{
  "mcpServers": {
    "skybuddy": {
      "command": "python",
      "args": ["scripts/mcp_server.py", "claude"],
      "cwd": "/path/to/SkyBuddy"
    }
  }
}

# the agent switches traveller with the user tools
create_user(user="ana", display_name="Ana")
switch_user(user="ana")
get_current_user()

# 1 · create the intent from a flight link
python scripts/booking_agent.py prepare --user harold \
  --booking-url "https://www.iberia.com/…" \
  --airline Iberia --origin BIO --destination BOG \
  --outbound-date 2026-12-04 --return-date 2027-01-08 \
  --price 684 --currency EUR --passenger harold --max-price 700

# 2 · approve it explicitly
python scripts/booking_agent.py confirm --intent-id bk_7f3a1c --approved-by harold

# 3 · hand the playbook to your agent
python scripts/booking_agent.py playbook --intent-id bk_7f3a1c

# or let the year of history decide and create the intent itself
python scripts/price_baseline.py trigger --origin BIO --destination BOG \
  --outbound-date 2026-12-04 --price 684 --airline Iberia \
  --booking-url "https://www.iberia.com/…" --passenger harold

MCP surface

Tools any agent can call

The same set is exposed to Claude, Hermes, OpenClaw and any MCP client through mcp_server.py. Every call runs inside the active traveller's workspace.

Tool	What it does
`create_user`	Create a traveller workspace with its own preferences, routes, bookings and history.
`list_users` / `get_current_user`	See every traveller and which one is active.
`switch_user` / `update_user` / `delete_user`	Change the active traveller, edit a profile, or remove one.
`search_flights`	Search a route and return fares plus the top scored recommendations.
`add_route`	Watch a route with an optional target price.
`list_routes`	List watched routes with their latest prices.
`check_all_routes`	Re-price every watched route and raise alerts.
`get_alerts`	Recent alerts within a time window.
`get_preferences` / `set_preferences`	Read and update airline, time, cabin and stop preferences.
`add_card` / `list_cards`	Manage credit cards used for points earning.
`add_loyalty_program`	Record a loyalty balance and tier.
`estimate_earnings`	Estimate points earned on a given fare.
`add_passenger` / `list_passengers`	Manage traveller profiles for bookings.
`prepare_booking`	Create a booking intent from a flight link, pending confirmation.
`confirm_booking`	Approve an intent after the price ceiling re-check.
`get_booking_playbook`	Return the step-by-step instructions the agent executes on the link.
`mark_booking_executed`	Record the outcome and confirmation code of a booking.
`list_bookings` / `cancel_booking`	Audit trail for every intent, and cancellation before execution.
`get_price_baseline`	Scan the last 365 days of recorded fares: min, percentiles, median, trend, confidence.
`evaluate_price`	Turn a live fare into a buy verdict with reasons and its percentile.
`auto_book_if_deal`	Create the booking intent automatically when the fare beats its own history.
`record_price_observation`	Feed one observed price into the baseline store.
`get_seat_advisory`	Seat-map workflow, cross-check sites and the actions that assign the seat.

Agent agnostic

Bring your own agent

Adapters ship in the box; anything speaking MCP works without one.

✓ Claude · MCP ✓ Hermes ✓ OpenClaw ✓ Any MCP client ✓ Plain Python / CLI

Open source

Track it. Score it. Book it.

Clone the repo, point your agent at the MCP server, and let SkyBuddy watch the fare while you sleep.

Open the repository Read the docs

Flight tracking youragent can act on

Tracked flights