Skip to main content

oddspush

Python client for the OddsPush API: real-time Pinnacle odds over one WebSocket, plus a REST API, with no-vig and fair prices, drop alerts, line history and the closing line for your CLV. Includes an MCP server for AI agents.

  • All 13 sports, live and pre-match.
  • A local price book, kept up to date for you: merged by version, closed markets and periods removed.
  • Reconnects on its own with backoff, subscribes again and catches up with a snapshot.
  • Sync and asyncio REST clients.
  • Pure Python, two dependencies (websockets, httpx), Python 3.10+.

You need an API key (opk_...). Sign up at oddspush.com and you get a free key at once (odds 30 seconds late, see Free plan); the 7-day real-time trial and the paid plans have everything. Full API reference: oddspush.com/docs.

pip install oddspush
pip install "oddspush[mcp]"     # with the MCP server, oddspush-mcp

Stream live odds

import asyncio
from oddspush import Stream

async def main():
    async with Stream("opk_...", sports=[1], channels=["ld", "both"]) as stream:
        async for msg in stream:
            if msg["type"] == "odds" and msg["data"].get("type") == "live":
                rec = msg["data"]["rec"]
                for key, market in stream.book.markets(rec["id"]).items():
                    print(rec["id"], key, [(p["designation"], p["price"]) for p in market["prices"]])

asyncio.run(main())

stream.book always holds the current open markets of every event you follow:

stream.book.event_ids()               # events in the book
stream.book.event(event_id)           # league, participants, state, startTime...
stream.book.markets(event_id)         # {"s;0;m": {...}, "s;0;ou;2.5": {...}}
stream.book.market(event_id, "s;0;m")

Prices are in American odds; oddspush.american_to_decimal(-155) gives 1.645.

Sports: 1 Soccer, 2 Tennis, 3 Basketball, 4 Hockey, 5 American football, 6 Baseball, 7 Rugby, 8 MMA, 9 Boxing, 10 Other (volleyball and handball), 11 Esports, 12 Golf, 13 Cricket (oddspush.SPORTS).

Channels: ld and both are the prices (ask for both). Add pre for pre-match prices. dz carries volatile moments: the book leaves it out, and with dz in your channels the stream asks for snapshots of the book's channels only. Without channels, live updates come on every channel; the book still takes only ld and both.

No-vig and fair prices

async with Stream("opk_...", sports=[1], channels=["ld", "both"], fair=True) as stream:
    async for msg in stream:
        if msg["type"] == "fair":
            for m in msg["data"]["markets"]:
                print(m["key"], m["margin"], [(p["designation"], p["fair_decimal"]) for p in m["prices"]])

Each fair message follows its odds update with the same markets priced without the bookmaker's margin, two ways: no_vig (multiplicative) and fair (power method, which corrects the favourite-longshot bias). The book keeps them too:

stream.book.fair(event_id, "s;0;m")   # {"version", "margin", "prices"} or None

It only returns the fair price of the very market version the book holds.

Closing line and CLV

from oddspush import Client

with Client("opk_...") as api:
    bet = api.clv(1639824805, "s;0;m", "home", 2.10)   # decimal, or American: -110, 150
    print(bet["clv_pct"], bet["beat_close"])           # vs the fair closing price, in %

    line = api.closing(1639824805)          # the whole closing line (pre-match or live id)
    for row in api.iter_closings(sport_id=1, since=1793900000000):   # up to 90 days back
        print(row["event_id"], row["start_ms"])

Closing lines are kept for 90 days after the start. Get them as they happen with Stream(..., closing=True) (closing messages), and work out the CLV of your own prices with oddspush.clv(your_decimal, fair_decimal).

Line history

Every change of price, line or status of an event's markets over the last 14 days, oldest first:

with Client("opk_...") as api:
    for row in api.iter_history(1639824805, key="s;0;m", fair=True):
        print(row["time"], row["channel"], row["status"],
              [(p["designation"], p["price"]) for p in row["prices"]])

history() returns one page (more, next_since); iter_history() follows the pages. The pre-match and the live version of a match are different events: ask for each.

Drop alerts

A drop message when a price shortens by at least min_pct % within window_s seconds, measured on the fair price, so a change of margin alone never alerts. Each alert has the match (league, teams, live score and minute), the market, the leg and the price before and after:

async with Stream("opk_...", sports=[1], drop_alerts={"min_pct": 5, "window_s": 60}) as stream:
    async for drop in stream.drops():
        print(drop["home"], drop["away"], drop["market"]["key"], drop["designation"],
              drop["from"]["decimal"], "->", drop["to"]["decimal"], drop["change_pct"])

drop_alerts=True means 3 % within 60 seconds; window_s is one of 30, 60, 300 or 600 and min_pct from 1 to 50. You can also pass on_drop=callback, or read drop messages from the main iterator. The setting is restored on reconnect like the rest of your filter. The drops of the last hour are on REST too:

with Client("opk_...") as api:
    page = api.drops(sport_id=1, min_pct=5, window_s=60)      # oldest first, with a cursor
    new = api.drops(sport_id=1, min_pct=5, window_s=60, after=page["cursor"])  # only new ones

Player props and specials

Specials (player, game and team props) are events of their own (rec.type == "special", parentId is the match). With specials=True your snapshots include them with their current prices, so the book recovers them after a reconnect:

async with Stream("opk_...", sports=[3, 5], channels=["ld", "both", "pre"],
                  specials=True) as stream:
    async for msg in stream:
        for event_id in stream.book:
            ev = stream.book.event(event_id)
            if ev.get("type") != "special":
                continue
            for key, m in stream.book.markets(event_id).items():
                names = {p["id"]: p.get("name") for p in ev.get("participants") or []}
                legs = [(names.get(p["participantId"]), p["price"]) for p in m["prices"]]
                print(ev["special"]["description"], legs,
                      "suspended" if stream.book.is_suspended(event_id, key) else "open")

Their markets have no status: a closed period suspends them (the book keeps them and is_suspended says so) and settled removes them.

REST

from oddspush import Client

with Client("opk_...") as api:
    api.events(sport_id=1)                          # every event, markets per channel
    api.event(1638807209, fair=True)                # one event, ld + both merged
    api.event(1639824805, channel="pre", fair=True) # a pre-match event
    api.membership(sport_id=1)                      # pre-match events about to go live
    for row in api.iter_closings(sport_id=1):       # matches with a closing line
        print(row["event_id"], row["start_ms"])

AsyncClient has the same methods, awaitable. A 429 is retried after retry-after (max_retries=2), then raised as RateLimited. Limits: 120 requests at once, then 300 a minute per account, times its connections with extras (20 and 60 on the free plan, per account and per IP).

Client(...).status() (or oddspush.get_status(), without a key) gives the public status: whether the feed is flowing and the latency measured every 5 minutes by a client connected like you.

Free plan

A free key gets the same odds, live and pre-match, all 13 sports, over the stream and REST, 30 seconds late: enough to explore and to build and test a bot, not to trade live odds. No fair prices, closing lines, CLV, line history, drop alerts or specials: those come with the 7-day trial and the paid plans.

You have to accept the delay, so that a bot never trades on late prices without knowing:

async with Stream("opk_...", sports=[1], accept_delay=True) as stream:
    async for msg in stream:
        if msg["type"] == "welcome":
            print("odds are", stream.delay_s, "seconds late")   # 30 free, 0 in real time

with Client("opk_...", accept_delay=True) as api:
    api.events(sport_id=1)
  • Without accept_delay=True a free key raises DelayedPlan (the stream stops, close 4110).
  • accept_delay=True is harmless on a paid key, and with it an expired key keeps receiving odds on the free plan instead of stopping with SubscriptionExpired.
  • If you upgrade while connected, the stream reconnects at once (plan_changed) and the odds arrive in real time.
  • A try_again (we could not check the key right now) is retried after retry_after_s.
  • Asking a free key for fair, closing, specials or drop_alerts logs a warning and they stay off (not_in_plan); on REST it raises NotInPlan.

MCP server for AI agents

pip install "oddspush[mcp]" adds oddspush-mcp, an MCP server (stdio) for Claude, Cursor and any MCP client that runs local servers. Its tools: list_sports, list_events, get_event, get_line_history, get_closing, list_closings, get_clv and recent_drops. Each tool calls the REST API from your machine with the key in ODDSPUSH_API_KEY.

claude mcp add oddspush --env ODDSPUSH_API_KEY=opk_... -- oddspush-mcp

Or, in a client's JSON config (Claude Desktop, Cursor...):

{"mcpServers": {"oddspush": {"command": "oddspush-mcp",
                             "env": {"ODDSPUSH_API_KEY": "opk_..."}}}}

With a free key, list_sports, list_events and get_event work, 30 seconds late; the other tools say which plans have them. Like any REST request, a call from a new IP beyond your connections takes the key over from the oldest IP: run it on the machine your bot runs on.

What the stream does for you

  • Logs in and subscribes. After a drop it reconnects with exponential backoff (1 s, 2 s, 4 s... up to 30 s), restores your filter and asks for a snapshot.
  • Asks for a snapshot after missed (you fell behind) and after feed: up (the source came back), and again later if the snapshot rate limit says so.
  • Yields {"type": "disconnected", "reason": ..., "retry_in_s": ...} each time the connection drops, so you can log it.
  • Snapshots count against your plan's snapshots_per_hour. Pass resync=False to stop the automatic ones.
  • Stops and raises on what retrying cannot fix: Unauthorized, SubscriptionExpired, NotInPlan (a REST-only plan), DelayedPlan (a free key without accept_delay=True) and TokenReplaced.

Change the subscription at any time:

await stream.subscribe(sports=[2, 3])
await stream.unsubscribe(sports=[3])
await stream.subscribe(fair=True)     # turn fair prices on, filter unchanged, no snapshot
await stream.subscribe(drop_alerts={"min_pct": 4})
await stream.unsubscribe(drop_alerts=True)
await stream.request_snapshot()

In a bot that runs for days, stream.book.prune(3600) drops events that have had no update for an hour and have no open market left.

One machine per connection

A token works from as many IPs at a time as the connections in your plan: one included, plus any extra connections you add from your account page (with your plan in one payment, ending with it, or on their own for 30 days; 90 USD a month each). Using it from a new IP beyond that (a stream login or a REST request) takes over from the oldest IP and closes the stream there (TokenReplaced). With one connection, run your stream and your REST calls from the same machine.

Keep up

Read fast and do heavy work in another task: if you fall more than 15 seconds behind, the server closes the connection and the stream reconnects. Host your bot close to Ireland (AWS eu-west-1) for the lowest latency.

Examples

See examples/: a live printer, a fair-price watcher, a pre-match tracker, drop alerts, a CLV report and a CLV report with the line history of each bet.

OddsPush is not affiliated with, endorsed by or sponsored by Pinnacle. MIT licensed.

Metadata

Release files for oddspush 0.3.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for oddspush 0.3.1
File Size Uploaded Provenance Actions
oddspush-0.3.1.tar.gz 65.2 kB With provenance Details for oddspush-0.3.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for oddspush 0.3.1
File Interpreter ABI Platform Actions
oddspush-0.3.1-py3-none-any.whl Python 3 none any Details for oddspush-0.3.1-py3-none-any.whl

Total release size: 113.4 kB

Release files / oddspush-0.3.1.tar.gz

Download URL oddspush-0.3.1.tar.gz
Size 65.2 kB
Tags Source
SHA-256 checksum
How to use checksums
57302eb981dc9ac61526d95e94d704ca3be46fcbca738b9a2774c90ebb1f3f6b
BLAKE2b-256 checksum
How to use checksums
b100c51e652785b2732ffcfdea4fa25bb6b4acbd77ca61265fc42850014c1290
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

Release files / oddspush-0.3.1-py3-none-any.whl

Download URL oddspush-0.3.1-py3-none-any.whl
Size 48.2 kB
Tags Python 3
SHA-256 checksum
How to use checksums
55cc7aa378d019acf0134eb889c1827804cf307a4bda542e61505ab200aab769
BLAKE2b-256 checksum
How to use checksums
f847298418bdb4d9e2659811cd7ffb49cc6ce68b5d63a7234ab7341d3ac2c09e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

Security for oddspush 0.3.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for oddspush-0.3.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 20, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page