Quick Start

# Install
pip install oddsharvester

# Or clone & setup with uv
git clone https://github.com/jordantete/OddsHarvester.git && cd OddsHarvester
pip install uv && uv sync

# Scrape today's upcoming football matches
oddsharvester upcoming -s football -d $(date +%Y%m%d) -m 1x2 --headless

# Scrape historical Premier League odds
oddsharvester historic -s football -l england-premier-league --season 2024-2025 -m 1x2 --headless

# Snapshot odds for matches in play right now
oddsharvester live -s tennis -m match_winner --headless

# Scrape community data (top predictions here; also --user profiles and --match-url votes)
oddsharvester community -s football --headless

# Find a team by name, then list its matches by the team_id it returns
oddsharvester search --query "Nacional Potosi" -s football --headless
oddsharvester search --team-id hUyau0Vc --headless

Features

Supported Sports & Markets

Umbrella tokens: pass one as --market and it expands at scrape time to every line OddsPortal renders for that match under the period asked for (e.g. over_under_1_5_market, over_under_2_5_market, …), instead of listing each line by hand. It opens only lines that have a token, and names the rendered lines it skipped in a warning. Outside football a whole line is written <n>_0 (over_under_games_161_0, asian_handicap_games_-5_0_games, handicap_+3_0), except the 0 line (handicap_0, asian_handicap_games_0_games).

Sport	Umbrella tokens
Football, basketball, American football, handball	over_under, asian_handicap
Ice hockey, baseball	over_under
Rugby league, rugby union	over_under, handicap

Handball's asian_handicap expands to handicap_* keys, and rugby's handicap umbrella is handicap, read from the Asian Handicap tab.

Tennis and volleyball lines: OddsPortal prints no Sets, Games or Points word on a line, and each axis's ladder can show any value in its range, so a line whose value falls within the other axis's range cannot be told apart on the page. Refused: every tennis sets Asian Handicap line; tennis games AH -2.5 and +2.5; tennis O/U sets and games from 6.5 to 10.5, whole games lines 7 to 10 included; volleyball sets and points AH -2.5, -1.5, +1.5, +2.5. They come back empty with a warning naming the other axis.

Cricket: the scraper supports one cricket market, home_away (a 2-way match winner, no draw), over one period, full_including_ot: the only ones OddsPortal shows on limited-overs matches (gotchas §14). Since the 2026-08 redesign the match pages carry a per-bookmaker odds table, read like any other sport's.

100+ leagues supported across all sports: Premier League, La Liga, Serie A, NBA, NFL, MLB, NHL, ATP/WTA Grand Slams, and many more.

CLI Usage

OddsHarvester has six commands: upcoming, historic, live, community, team and search. The CLI Options Reference gives, for each option, the commands that take it.

oddsharvester upcoming

Scrape odds for upcoming matches — by date, by league, or by specific match URL.

# By date (today here)
oddsharvester upcoming -s football -d $(date +%Y%m%d) -m 1x2 --headless

# By league (scrapes all upcoming matches for that league)
oddsharvester upcoming -s football -l england-premier-league -m 1x2,btts --headless

# Multiple leagues
oddsharvester upcoming -s football -l england-premier-league,spain-laliga -m 1x2 --headless

# Specific match URLs (repeat the flag; works for past matches too)
oddsharvester upcoming -s football --match-link "https://www.oddsportal.com/football/..." -m 1x2

# Preview mode (faster — best/highest odds only, no individual bookmakers)
oddsharvester upcoming -s football -d $(date +%Y%m%d) -m over_under --preview-only --headless

# Only matches kicking off within the next 6 hours (fewer requests)
oddsharvester upcoming -s football -l england-premier-league -m 1x2 --kickoff-within-hours 6 --headless

# Collect links plus kickoff only (fixture plan; no odds scraped)
oddsharvester upcoming -s football -d $(date +%Y%m%d) --links-only -f csv -o upcoming_links.csv

oddsharvester historic

Scrape historical odds and results for past seasons.

# Single league & season
oddsharvester historic -s football -l england-premier-league --season 2022-2023 -m 1x2 --headless

# Current season
oddsharvester historic -s football -l england-premier-league --season current -m 1x2 --headless

# Limit pagination
oddsharvester historic -s football -l england-premier-league --season 2022-2023 -m 1x2 --max-pages 3 --headless

# Output as CSV
oddsharvester historic -s football -l england-premier-league --season 2024-2025 -m 1x2 -f csv -o premier_league_odds --headless

# Umbrella market — expands to every Over/Under line rendered on the page
oddsharvester historic -s football -l england-premier-league --season 2023-2024 --market over_under -f csv

oddsharvester live

Take a one-shot snapshot of matches currently in play, with per-bookmaker in-play odds.

# Every live match for a sport
oddsharvester live -s tennis -m match_winner --headless

# Restricted to one league
oddsharvester live -s football -l england-premier-league -m 1x2 --headless

# Re-sample one known match without reloading the listing
oddsharvester live -s football --match-link "https://www.oddsportal.com/football/..." -m 1x2 --headless

# Collect live match links only
oddsharvester live -s football --links-only -o live_links.json --headless

Each record carries the usual match metadata plus live context: live_period (the period marker exactly as the site displays it, so it is sport-specific: 4' or Half-time in football, 1st Set in tennis, 9th Inning in baseball), live_score_home, live_score_away, live_score_raw (keeps compound formats such as 0:1 (3:6, 4:2)), and scraped_at_utc, which is what makes a series of snapshots comparable.

Notes:

• Zero live matches is a normal outcome: the command prints a message and exits 0 without writing a file. upcoming and historic exit 1 on a run that scraped no match at all, an empty listing included.
• Matches that end between listing and scrape are dropped, so a snapshot only ever contains genuinely live matches.
• In-play bookmaker coverage is thinner than pre-match (often 2 to 4 bookmakers instead of 15 to 20) and varies by region.
• --odds-history is rejected: the in-play view has no history. --period takes any period of the sport and reads that period's in-play odds; a period the in-play view does not offer for a market leaves that market empty.
• For repeated sampling, schedule the command externally (cron or similar). Keep at least 60 seconds between snapshots, and prefer --match-link to re-sample a known match without re-reading the listing. The command itself never refreshes a page, which keeps its request profile identical to upcoming.

oddsharvester community

Scrape OddsPortal Community data. community has three mutually-exclusive modes; exactly one is required:

• Top predictions (--sport): the most-voted community picks for the next 7 days.
• User profile (--user <username>): a tipster's stats, monthly performance and recent predictions.
• Match community votes (--match-url <url>): the community vote split of one match, for the market its page shows on load.

Top predictions (--sport)

# Top predictions for a sport
oddsharvester community -s football --headless

# Write to a named JSON file
oddsharvester community -s football -f json -o top_predictions.json --headless

Each record contains the match (home_team, away_team, match_url, kickoff as YYYY-MM-DDTHH:MM in the browser timezone (--timezone), plus the raw kickoff_text label kept as fallback when the date token fails to parse), the league (sport, country, league), the voted market, best odds per outcome (odds), the community vote split (community_votes_pct), and scraped_at.

• OddsPortal surfaces ~10 picks per sport (no pagination) with rounded percentages.

User profile (--user <username>)

oddsharvester community --user BLAPRO --headless

Emits one record: mode ("user"), the header (username, roi_pct, member_since, country, privacy), the monthly statistics table (month, total_predictions, won, lost, plus_minus, roi_pct, incl. a Total row), the rendered predictions batch and scraped_at. Each prediction has kickoff (YYYY-MM-DDTHH:MM in the browser timezone) with the raw kickoff_text kept as fallback, market, home_team/away_team, score (when finished), match_url, and a positional outcomes list of {odds, community_pct, picked} plus pick_odds. Most profiles are private: a private profile returns the header only (privacy: "private", empty stats/predictions) and exits 0.

Match community votes (--match-url <url>)

oddsharvester community --match-url "https://www.oddsportal.com/football/h2h/.../" --headless

Emits one record: mode ("match"), match_url, event_id (the id after the URL's #), home_team, away_team, kickoff, is_prematch, markets and scraped_at. markets holds the market the page shows on load (1X2 for a football URL ending in #<id>) as {market, scope, outcomes}, each outcome {outcome, votes_pct}: the label of its odds column (1, X, 2) and its rounded share of the votes. A finished match can keep its votes: is_prematch is false once the header shows the live marker or a score. When the page shows no vote row, markets is empty and the command exits 1. Its kickoff reads like the match header, weekday first (Sunday, 04 Jan 2026, 17:30), in the browser timezone (--timezone).

Limitations: --match-url reads one market, the one shown on load, and only as percentages, since the page no longer carries vote counts (gotchas §19). --user captures the first rendered predictions batch (no deep pagination) and does not emit per-prediction win/loss (use the monthly stats table).

oddsharvester team

Scrape team metadata from OddsPortal team pages: identity, venue, coach and recent form. One record per team.

# By team id, as it appears in an OddsPortal URL
oddsharvester team --team lId4TMwf --headless

# Several teams at once, by id or by pasted team page URL, into a spreadsheet
oddsharvester team --team lId4TMwf,WGt8En5I --headless -f csv -o teams.csv

# From a file holding one id or URL per line
oddsharvester team --teams-file my_teams.txt --headless -o teams.json

Each record carries team_id, name, full_name, list_name, sport, country, town, venue, coach, tournament, logo_url, the Last 6 Games block (form, over_2_5_pct, btts_pct, avg_goals_scored, avg_goals_conceded), the canonical team_url and scraped_at.

list_name is the short name OddsPortal prints in match lists, which is the one to join on: a team can appear as "Barcelona FC" on its own page and "Barcelona" in a fixture list. It is read from the recent-form block, so it survives a page that shows no odds rows at all.

Limitations: country, town and venue are simply absent for smaller teams, and tournament only appears on fixture rows, which your IP's selected bookmakers may leave empty. A wrong team id is reported as a failure rather than written as a blank record: the page it returns builds its heading from the URL and otherwise looks valid.

Team pages are loaded one after the other, --request-delay seconds apart (default 1.0), so a run of N teams takes about N seconds more than the pages themselves.

oddsharvester search

Find a team without knowing its league, then list its matches. OddsPortal's search runs in two steps, one call each.

# 1. The teams whose name matches, within one sport
oddsharvester search --query "Nacional" --sport football --headless -o teams.json

# 2. One team's matches, by the team_id step 1 returned (or any team id or team page URL)
oddsharvester search --team-id hUyau0Vc --headless -o matches.json

# Further back in its history: 20 matches per results page
oddsharvester search --team-id hUyau0Vc --max-pages 3 --headless -o matches.json

A --query row carries team_id, name, country, sport and team_url, most relevant first; a common name returns several teams (19 for "Nacional"), so pick by country. A --team-id row carries match_link and kickoff_utc, in the same shape as --links-only rows, then tab (next or results), home_team, home_team_id, away_team, away_team_id, home_score and away_score (empty for a match to come), partial_results (the period scores, as on historic rows), tournament, league_url, country and sport. Upcoming matches come first, then results, newest first. The scores are the ones OddsPortal shows: after a penalty shoot-out the winner gets one more goal, so a 1:1 draw won on penalties reads 2:1; partial_results ((0:1, 1:0, 0:0, 4:3)) holds the shoot-out as its last pair. In tennis, here and in historic, a lost tie-break's points follow the loser's games: (6:3, 6 3 :7, 6:3) is a 6:7 second set whose tie-break the home player lost with 3 points, and 7:6 4 one the away player lost with 4.

The command exits 0 whenever the pages were read, and writes [] when nothing matched. It exits 1 on a refused page (rate limited by OddsPortal), a team id the site does not know (does not exist on OddsPortal), any page that fails, or an output it cannot write. A --team-id call that loses one page writes nothing rather than a partial list.

Limitations: the upcoming tab lists only the matches your IP's selected bookmakers price, so a small club's next match can be missing until its odds open; it shows in the results once played. Only the first page of upcoming matches is read.

CLI Options Reference

The Commands column names the commands that take each option; all stands for all six.

Global Options

Given before the command name, e.g. oddsharvester -v upcoming ....

Core Options

--match-link usage: --sport is still required. Match links bypass the listing pages entirely, so upcoming and historic scrape any match, played or not. historic takes an optional --season (one at most) and writes it into each record's season field, which the match page does not give. For large link sets (a --links-only output, re-running failures), prefer --match-links-file: a pasted command line gets silently truncated by the terminal past a few hundred URLs.

upcoming only: --date is required unless --league or --match-link is provided. --date and --league can be combined to filter the league's upcoming matches down to a specific calendar day. When combining both, the reference timezone for resolving the date is --timezone if provided, otherwise the host's timezone. --kickoff-within-hours N keeps only matches starting within N hours from now; the filter runs during link collection, so far-off matches are never visited. It pairs with the default upcoming-only behaviour to bound the window on both sides; the window is counted in real hours from now, whatever the timezone. Combined with --links-only, each row also carries kickoff_utc, so a scheduler can plan a day of fixtures from one listing request instead of re-fetching the listing on every cycle.

historic only (--max-pages also applies to search):

historic needs --league or --match-link; without either it exits 2 before the browser starts.

live only: no --date and no --season; the command always reads whatever is in play at the moment it runs. --league accepts at most one slug. --odds-history is rejected, because the in-play view exposes no history; --period selects the period in the in-play view's own period bar. --links-only cannot be combined with --local-kickoff, as on upcoming and historic. A --match-link given in classic form is normalized to its in-play URL automatically, so either form works. When every match fails to scrape the command exits non-zero, which is what lets a scheduled sampler tell a blocked run apart from a genuinely empty one.

Community, Team and Search Options

community takes exactly one of --sport, --user and --match-url; team needs --team or --teams-file; search takes exactly one of --query and --team-id. All three exit 2 before the browser starts otherwise.

Output Options

Breaking change: a run now exits 1 when its output cannot be written. The batch is then saved next to the requested output as <name>.unsaved-<UTC timestamp>.json, and the message gives that path. --append onto an existing JSON file that is not a readable list (invalid JSON, which used to be overwritten, or JSON that is not a list, which used to fail silently) leaves that file as it was and sends the batch to the fallback file. JSON output, and CSV output without --append, is written to a temporary file first and then moved into place, so an interrupted run never leaves a truncated file. When the output's folder does not allow a temporary file, the output is written in place as before. historic and upcoming also exit 1 when a league or season fails to list, after writing the data of the others.

Breaking change: every output row now carries a season column. For odds rows it is inserted directly after match_date; --links-only rows have no match_date and carry season alongside the other link fields instead. It holds the scraped season for historic, including the --season given with historic --match-link, and is empty for upcoming and other --match-link runs. Appending to a CSV file produced by an earlier version widens its header: earlier rows get an empty season.

Breaking change: upcoming --links-only rows now carry a kickoff_utc column, appended at the end. Appending to a CSV file produced by an earlier version widens its header: earlier rows get an empty kickoff_utc. live links-only rows are unchanged; historic ones gain match_day (below).

Breaking change: historic --links-only rows now carry a match_day column, appended at the end. Appending to a CSV file produced by an earlier version widens its header: earlier rows get an empty match_day. upcoming and live links-only rows are unchanged.

Breaking change: boto3 moved to the s3 extra, so pip install oddsharvester no longer installs it, and there is no default S3 bucket any more. --storage remote (or OH_STORAGE=remote) needs pip install 'oddsharvester[s3]' and a bucket in OH_S3_BUCKET; without either, the command exits 2 before the browser starts and names the missing piece. Remote storage now honours --format and --append the same way local storage does (it used to always write JSON and always overwrite), and the object key carries the file's extension when -o has none, e.g. -o out now uploads out.json.

Browser & Scraping Options

Proxy Options

League listings always load on the first proxy's browser context; only match pages rotate across the configured proxies.

Tip: For best results, match --locale and --timezone to your proxy's region.

Multi-proxy example — spread scraping across three proxies with embedded credentials:

oddsharvester historic --sport football --league england-premier-league --season 2013-2014 \
  --market 1x2 --concurrency 6 \
  --proxy-url http://user:pass@p1.example.com:8000 \
  --proxy-url http://user:pass@p2.example.com:8000 \
  --proxy-url http://user:pass@p3.example.com:8000

Matches are dispatched round-robin across the proxies; a proxy that fails 3 times in a row (navigation/rate-limit errors) is dropped from rotation and the run continues on the survivors.

Advanced Options

Deprecated: --odds-format never changed anything: odds are always decimal. It is hidden from --help, a non-decimal value prints a warning, and it will be removed in a future release.

Breaking change: --odds-history timestamps now carry the match's year (the year before kickoff for a December opening of a January match) instead of the year the scraper runs in. They stay naive (no UTC offset) and are in the browser timezone: --timezone when given, otherwise the host's. Every bookmaker entry gets exactly one odds_history_data block per outcome, in outcome order; a block whose history cannot be read is {"odds_history": [], "opening_odds": null} instead of being dropped, which used to shift the remaining blocks. A market requested for a non-default --period that cannot be verified on the page is now returned empty instead of carrying another period's odds. Every period is selected through its URL scope code, on www.oddsportal.com and on regional mirrors alike, and a non-default period is kept only when the page shows that period's tab (a later tab is bold, or the first tab, once clicked, writes that period's scope): a period the match does not offer (for example NFL 2nd_half or baseball full_time on a match without that tab) comes back empty instead of carrying the default period's odds.

Breaking change: the sport's default period is checked too, against the periods the match's data list bookmakers under for the market. A market without it comes back empty, with a warning naming the --period values it has, instead of carrying another period's odds under the default's label: basketball and American football 1x2 (listed under Full Time only, while their default is full_including_ot), or an ice hockey home_away listed under FT including OT only. Ice hockey now takes --period full_including_ot, basketball and American football --period full_time.

Two-pass workflow: collect links, then scrape

For large runs it can be safer to collect all match links first, then scrape odds per link and re-run only the failures (see issue #75):

# Pass 1 - collect the season's match links (no odds scraped)
oddsharvester historic -s football -l england-premier-league --season 2022-2023 \
    --links-only -f csv -o links.csv

# Pass 2 - scrape odds from the collected links (--append fills recovered failures)
tail -n +2 links.csv | cut -d, -f1 | sort -u > links.txt
oddsharvester upcoming -s football -m 1x2 -f csv -o odds.csv --append \
    --match-links-file links.txt

On upcoming, the same pass 1 doubles as a fixture plan: one listing request returns every match of the day with its kickoff, so a scheduler can decide offline which matches are close enough to be worth scraping.

# Plan the day's fixtures: links plus kickoff, no odds scraped
oddsharvester upcoming -s football -d $(date +%Y%m%d) --links-only -f csv -o upcoming_links.csv

Output rows contain match_link, sport, league, and season (date and kickoff_utc for upcoming, match_day for historic; live emits none of them), in the site's listing order. match_day (2025-05-25) is the day of the date header the results page groups the match under, as the browser renders it, so in the --timezone zone (a results page shows no kickoff time), and is empty for a row with no readable date header above it. kickoff_utc holds the match's own kickoff in UTC, in the same shape as match_date (2026-07-31 18:30:00 UTC), and is empty when the listing exposes no parseable kickoff. Under the default upcoming-only behaviour that means the date header could not be read or the row's first column held no HH:MM clock; with --include-started, a row whose first column shows a status or a period marker instead of the clock (a live or finished match) also comes back with an empty kickoff_utc. Options that only affect odds scraping (--market, --period, --odds-history, --preview-only, --target-bookmaker, --bookies-filter) are ignored when --links-only is set. --links-only cannot be combined with --match-link.

Bulk scraping: multiple leagues, multiple seasons

--season and --league both accept comma-separated lists. historic scrapes every combination as the cartesian product, league outer and season inner; listings run up to --concurrency at once, but output still stays grouped and deterministic per combo. --max-pages applies per combo, not per run.

# Several seasons of one league
oddsharvester historic --sport football --league england-premier-league \
    --season 2020-2021,2021-2022,2022-2023 --links-only --format csv --output links.csv

# Cartesian product: every league by every season
oddsharvester historic --sport football \
    --league england-premier-league,spain-laliga \
    --season 2021-2022,2022-2023 --links-only

# A league that changed season format mid-history: pass both formats and
# let the invalid pairs report zero
oddsharvester historic --sport football --league russia-premier-league \
    --season 2010,2010-2011,2011,2011-2012 --links-only

No pre-filtering is attempted to figure out which (league, season) pairs are valid before scraping them; a wrong-format pair (e.g. 2010 for a league that only ever used 2010-2011) is simply scraped and returns zero links. That is normal, not an error: OddsPortal returns HTTP 200 for a dead season URL, so the only way to tell a valid combo from an invalid one is to scrape it and count the results (see docs/agentic-gotchas.md §15). When more than one combo runs, an end-of-run table lists each league/season pair with its count, then reports how many combos returned nothing and how many errored. A zero-count combo does not fail the run while another combo returns data; when every combo returns nothing, the run exits 1, as any run that scraped no match does. An errored combo makes the run exit 1 once the other combos' data is written, and is the only kind worth re-running.

Local kickoff time

--local-kickoff adds two fields to each record: venue_timezone (the venue's IANA timezone id) and match_date_venue_local (the kickoff converted to that timezone, with an explicit offset), e.g. 2022-05-01 16:00:00 BST+0100. match_date stays UTC; the two fields are additive and only appear when the flag is set.

Resolution is best-effort from the record's venue country/town. Single-timezone countries resolve by country; USA, Canada, Mexico, Brazil, Russia, and Australia resolve by host city instead. A venue that can't be resolved gets null for both fields.

Not compatible with --links-only (no match pages are visited, so there's no venue to resolve). Distinct from --timezone, which sets the browser's context timezone (the host's when unset): a league listing groups its dates in that zone, so it decides which matches -d keeps there, and the community kickoff fields and the --odds-history timestamps are given in it.

Appending to an existing CSV file adds the two columns to its header; rows written before carry them empty.

Streaming results while the run is in progress

By default a run collects every match and writes the whole batch at the end, so a 10-match run hands you nothing for 10 minutes. --stream-ndjson emits each match on stdout as a single JSON line the moment that match is done, letting a downstream process start work while the rest is still being scraped:

oddsharvester upcoming -s football -d $(date +%Y%m%d) -m 1x2 --stream-ndjson --headless \
    | while read -r line; do echo "$line" | jq -r '.home_team + " vs " + .away_team'; done

Each line is exactly the record that --output would have written, so a consumer parses the same shape either way. Logs, the run summary and the failed-URL list all go to stderr, leaving stdout carrying nothing but NDJSON.

Without --output, no result file is written at all — the stream is the output. Pass --output to keep a batch file as well; --format then governs that file only, the stream stays NDJSON.

Failures are not streamed: they stay visible on stderr and in the exit code. Available on upcoming, historic and live; not compatible with --links-only, which never visits match pages. A run that spans several leagues or seasons streams every match into one flat sequence with no marker between combos.

Uploading to S3

--storage remote writes the output file exactly as --storage local does, then uploads that file to an S3 bucket:

pip install 'oddsharvester[s3]'
export OH_S3_BUCKET=my-odds-bucket
export OH_AWS_REGION=eu-west-1   # optional, default eu-west-3
oddsharvester historic -s football -l england-premier-league --season 2024-2025 -m 1x2 \
    --storage remote -f csv -o data/epl.csv

The object key is the local path as written: data/epl.csv above, scraped_data.json without --output. The local file stays on disk, and --format and --append apply to it as they do locally, so each run uploads the whole file. Credentials come from the usual AWS sources (environment variables, ~/.aws/credentials, an instance role). When the upload fails, the command exits 1 and names both the local file that holds the records and the s3:// location it could not reach. With --stream-ndjson, remote storage needs --output, since the stream alone writes no file to upload.

Blocked odds

OddsPortal strikes through a price when that bookmaker has stopped offering the bet. Each per-bookmaker odds record then carries a blocked_outcomes field listing which outcomes are struck through, using the same labels as the odds themselves:

{
  "bookmaker_name": "Unibet.fr",
  "1": "1.32",
  "X": "4.55",
  "2": "6.10",
  "blocked_outcomes": ["1", "X", "2"]
}

The labels are the market's own, so they differ per market: 1 / X / 2 for 1X2, odds_over / odds_under for Over/Under, and so on.

The field is always collected, with no flag to enable, and is omitted entirely when nothing is blocked, so records for available odds are unchanged. Odds values are kept exactly as rendered: a struck-through price is still the last price that bookmaker showed. A bookmaker with no price at all renders - and is not flagged, so "no odds" and "blocked" stay distinguishable.

Environment Variables

The options in the table below can also be set through an environment variable, which is handy for Docker or CI/CD; a value given on the command line wins over the variable. The other options have no variable and are set on the command line only.

export OH_SPORT=football
export OH_HEADLESS=true
export OH_PROXY_URL=http://proxy.example.com:8080

oddsharvester upcoming -d $(date +%Y%m%d) -m 1x2

Installation

With pip (from PyPI)

pip install oddsharvester

# With S3 upload (--storage remote)
pip install 'oddsharvester[s3]'

From source (with uv)

git clone https://github.com/jordantete/OddsHarvester.git
cd OddsHarvester
pip install uv
uv sync

python3 -m venv .venv
source .venv/bin/activate    # Unix/macOS
# .venv\Scripts\activate     # Windows

pip install . --use-pep517
# or: poetry install

Browser

Playwright downloads its Chromium separately, once per machine:

playwright install chromium          # after pip install
uv run playwright install chromium   # from source

On a Linux server, add --with-deps to install the system libraries the browser needs too (playwright install --with-deps chromium). Runs use Playwright's full Chromium build, headless or not: an install made with --only-shell is not enough.

Verify installation:

oddsharvester --help

Docker

# Build
docker build -t odds-harvester:local .

# Run (CLI args are appended to the ENTRYPOINT `python3 -m oddsharvester`)
docker run --rm odds-harvester:local upcoming -s football -d $(date +%Y%m%d) -m 1x2 --headless

# Run and keep the JSON output on the host (mount a volume + use -o)
# On macOS+colima, prefer a path under $HOME (e.g. $PWD); /tmp is not shared by default.
docker run --rm -v "$PWD/_docker_out:/out" odds-harvester:local \
  upcoming -s football -d $(date +%Y%m%d) -m 1x2 --headless -o /out/result.json

# Or with environment variables
docker run --rm \
  -e OH_SPORT=football \
  -e OH_HEADLESS=true \
  odds-harvester:local upcoming -d $(date +%Y%m%d) -m 1x2

Contributing

Contributions are welcome! Submit an issue or pull request. Please follow the project's coding standards and include clear descriptions for any changes.

License

MIT License

Disclaimer

This package is intended for educational purposes only. The author is not affiliated with or endorsed by oddsportal.com. Use responsibly and ensure compliance with their terms of service and applicable laws.