---
title: "Score Prob"
method: POST
path: "/v1/mcp/score_prob"
tags: ["mcp"]
---

# Score Prob

`POST /v1/mcp/score_prob`

## Request

- Base URL: `https://api.infersports.dev` (the document declares no server; this is the origin it was published from)
- URL: `https://api.infersports.dev/v1/mcp/score_prob`
- Auth: none declared

## Request body

- ScoreProbRequest
  - `query` string, required: Natural-language fixture, e.g. 'Mexico vs South Africa' or a team.
  - `sport` string, nullable
  - `date` string, nullable
  - `top` integer: How many scorelines to return (1-10).

## Response `200`

Successful Response

- ScoreProbResult: Market-implied correct-score probabilities for ONE football match. The de-vigged 1x2 + totals markets are inverted into a bivariate-Poisson (Dixon-Coles) score grid; the asian-handicap fair price is held out and reported as ``ah_ev_check`` (≈0 = the markets agree with the grid). When the moneyline is suspended (common once a match is decided) but the books keep totals + AH open, supremacy is solved from the handicap instead and ``ah_ev_check`` is null. Live matches are conditioned on the current score: the lambdas describe REMAINING goals and scorelines are still full-time (never below the current score). This is what the market implies, not a house prediction. ``status`` mirrors get_sharp_line: ok | ambiguous | not_found | no_data.
  - `status` string, required
  - `query` string, required
  - `summary` string, required
  - `match` MatchSummary: A fixture flattened for agent consumption (teams as names, optional live score).
    - `event_id` string, required: Canonical event id (evt_…); pass to get_match_odds / compare_lines.
    - `sport` string, required: football | basketball
    - `league` string, required: League name, e.g. 'English Premier League'.
    - `home_team` string, required
    - `away_team` string, required
    - `scheduled_at` string, nullable: Kickoff, ISO-8601 UTC.
    - `scheduled_at_local` string, nullable: Kickoff in the requested timezone (ISO-8601); null when no timezone was given.
    - `status` string, required: scheduled | live | finished
    - `score` Score
      - `home` integer, required
      - `away` integer, required
    - `red_cards` Score
      - `home` integer, required
      - `away` integer, required
    - `clock` string, nullable: Match clock when live, humanized for display: e.g. '1st half 25'', '2nd half 47'' (the real minute, not estimated) or 'half-time'; null pre-match or when the feed omits it. Basketball clocks (e.g. 'Q2 4:12') pass through as-is.
    - `confidence` number, nullable: Fuzzy-match confidence 0–1 (find_match only; null elsewhere).
    - `summary` string, nullable: One-line human-readable status, e.g. "Man City vs Arsenal — LIVE 1h 23 2-1." or "… — Kicks off 18:00 UTC on Jun 01." — ready to read to a user.
  - `scores` ScoreEntry[]: Top-K scorelines by probability, descending.
    - `score` string, required: Full-time scoreline home-away, e.g. '1-0'.
    - `prob` number, required: Market-implied probability of that exact score.
  - `top_cum` number, nullable: Probability mass covered by the returned scores.
  - `implied_1x2` object, nullable: The grid's home/draw/away probabilities — the de-vigged 1x2 it was fit to, or, when the moneyline was suspended and supremacy came from the asian handicap, the 1x2 the fitted grid implies.
  - `lambda_home` number, nullable: Fitted Poisson goal rate, home (remaining goals when the match is live).
  - `lambda_away` number, nullable: Fitted Poisson goal rate, away (remaining goals when the match is live).
  - `rho` number, nullable: Dixon-Coles low-score adjustment.
  - `ah_ev_check` number, nullable: EV of the fair asian-handicap price under the fitted grid (held-out cross-check; ≈0 means 1x2/totals/AH are internally consistent).
  - `fair_from` string, nullable: Bookmaker whose de-vigged 1x2 anchored the fit.
  - `as_of` string, nullable
  - `stale` boolean, nullable
  - `alternatives` MatchAlternative[]
    - `event_id` string, required
    - `label` string, required: 'Home vs Away' label for disambiguation.
    - `confidence` number, required
  - `ask_user` string, nullable
  - `decision` Decision, required: Pre-computed go/no-go for the caller (W3): we decide whether it's safe to proceed so the agent doesn't have to reason it out (and risk guessing). Branch on ``safe_to_proceed``: true → act on the result without asking; false → put ``ask_user`` to the human, or follow ``next_action`` when there's a clear recovery step (e.g. nothing resolved → list fixtures).
    - `safe_to_proceed` boolean, required: True when the result is unambiguous enough to act on without asking the user.
    - `confidence` number, nullable: Resolution confidence 0–1 (null when not applicable).
    - `ask_user` string, nullable: A disambiguation question to put to the user; null when safe_to_proceed.
    - `next_action` NextAction: The single best next call to make, pre-computed so the agent doesn't have to plan it (W3).
      - `tool` string, required: Tool to call next, e.g. 'get_match_odds' | 'compare_lines' | 'list_today_matches'.
      - `args` object: Ready-to-use arguments for that call, e.g. {'event_id': 'evt_…'}.
      - `reason` string, required: Why this is the suggested next step.

## Other responses

- `422`: Validation Error

---

[API](https://skmtc.dev/infersports/apis/infersports-api.md) · [All operations](https://skmtc.dev/infersports/apis/infersports-api/llms.txt) · [OpenAPI document](https://skmtc.dev/infersports/apis/infersports-api/revisions/1a1c3b357ea7?raw)