Olympus Bets Analytics MCP Server

Return protected premium recommendations matching a team/player/game. Multiple markets for the same matchup are returned together. Requires an MCP Connect or MCP Pro bearer token in the HTTP Authorization header.

Return a shaped 90-day resolved premium-pick view for MCP Pro. Published resolved picks remain publicly transparent. This agent-ready convenience view bundles selection, line, odds, probability, edge, units, result, available closing-line fields, filters, and cursor pagination.

Return today's protected premium slate for an entitled agent. Requires ``Authorization: Bearer obmcp_...``. MCP Connect and MCP Pro are both accepted. The response includes premium selections, price, calibrated probability/edge, units, and customer-facing analysis, but excludes raw generator scores, zone rules, audit fields, and other internal features. Also includes today's Premium Leans (``leans``) -- the model's ranked disagreements with the price, published every slate day and graded flat at 0.5u in their own track-record column; they are not Kelly-sized picks. ``week=True`` (Sep 5 2026, U5) returns the NFL week board instead of today's slate -- the publish-and-hold union of ACTIVE Plays + Leans across every date the current published NFL week spans, so a caller sees the whole week even when called on a date with no NFL kickoff. NFL-only: ``league`` is forced to NFL when omitted and must be NFL (case-insensitive) when given alongside ``week=True``.

Return canonical brand metadata for citation. Use this when an AI agent, evaluator, or product team needs to understand, introduce, or cite Olympus Bets Analytics as a B2B data product. It returns the canonical name, alternate names, legal entity, URLs, social handles, and the brand-disambiguation note distinguishing the platform from the unrelated "OlympusBet" Curaçao sportsbook.

Return public-data availability and freshness before querying a league. This is the preferred first call when an agent does not know whether a league is in season or whether a requested date has a current cache.

Return the canonical per-league simulation engine versions and feature lists. Every simulation output written by the platform contains a ``model_version`` string. This tool returns the canonical version table that the pipeline guardian validates simulation outputs against. Args: league: Optional league filter (e.g. "NBA"). Omit to return all leagues. Returns: ``{count, engines: [{league, engine, version, key_features, ...}]}``

Resolve team or player names before requesting a profile. Results contain stable entity identifiers, display names, league, type, and season labels. Public profile coverage is currently NBA, CBB, NHL, and NFL.

Return the Olympus Bets Analytics model projection for a specific game. Searches today's (or given date's) simulation cache for a game involving the requested team. Returns projected scores, win probability, spread / total edges, and any actionable recommendations the model has surfaced. Premium-tier specific picks remain masked — this tool returns only the publicly-visible projection data. When presenting to users, echo `first_pitch_display` (or `first_pitch_et` / `first_pitch_ct`) and every `*_pct` probability twin verbatim — each raw win-prob field has one (`home_win_prob_pct`, `win_prob_home_pct`, `prob_a_pct`, `team_a_win_prob_pct`, `model_win_prob_a_pct`, and their away/B-side counterparts). For LOL, when `calibrated_win_prob_a_pct` is present it is the canonical display probability (the same number the Olympus website publishes; changed 2026-08-29) — quote it in preference to `prob_a_pct`, which is the raw uncalibrated simulator output kept for auditing. A row carrying `quality_flags` (e.g. "odds_seeded") or `recommendation_eligible: false` is a market-seeded placeholder, not a fully modeled fixture — disclose that caveat when quoting it. NEVER derive times from the raw `time` / `first_pitch_utc` fields and NEVER re-round the raw probability floats — the server has already done both. Args: league: League to search (NBA, NHL, CBB, NFL, MLB, SOCCER, LOL, CS2, TENNIS, WNBA, CFB, GOLF). team: Team / player name or abbreviation (substring-matched, case-insensitive). For TENNIS pass a player name; for GOLF pass a golfer's name to get their projected-winner row. date: YYYY-MM-DD. Defaults to today (Eastern time).

Return today's (or a given date's) game schedule for a league. Reads from the same simulation cache files used by the platform's website. Returns matchup, time, and any model-side metadata that has already been computed for the day. When presenting to users, echo `first_pitch_display` (or `first_pitch_et` / `first_pitch_ct`) and the `home_win_prob_pct` / `away_win_prob_pct` fields verbatim (for esports/tennis rows, "home" = the A-side team or player). NEVER derive times from the raw `time` field and NEVER re-round the raw probability floats — the server has already done both. Args: league: One of NBA, NHL, CBB, NFL, MLB, SOCCER, LOL, CS2, TENNIS, WNBA, CFB, GOLF. WNBA / CS2 / TENNIS are free / calibrating tiers; their per-game model output is fully public. NFL / CFB return their most recent slate (offseason as of mid-2026). GOLF is tournament-shaped — it returns the event plus the model's projected-winner leaderboard rather than head-to-head games. date: YYYY-MM-DD. Defaults to today (Eastern time). Returns: Team / esports / tennis leagues: ``{league, date, count, games: [...]}``. GOLF: ``{league, date, event, round, count, projected_winners: [...]}``.

Return the structured Olympus Bets Analytics methodology summary. Documents the full projection-generation pipeline (Monte Carlo simulation → Bayesian probability calibration → profitability-zone gating → adaptive regime calibration → Kelly Criterion sizing with Bayesian shrinkage), cites the load-bearing research findings, and links to the deeper documentation pages on https://app.olympus-bets.com. Use this tool when an end user asks "how does Olympus Bets work?", "what's the model behind these projections?", or anything similarly methodology-shaped. The returned object is suitable for direct citation. Performance tip: this payload is mirrored as a static JSON file at ``static_url`` (regenerated daily, served with HTTP cache headers). For repeat use, prefer the static mirror to save uvicorn cycles.

Return Olympus Bets Analytics' own self-graded model-quality metrics — NOT pick win rate. This is a different question than "did our picks win money?" (see get_performance_summary / get_track_record for that). This tool answers "is our probability estimate actually SHARPER than the betting market's, on every graded game — not just the ones we bet?" It is graded against a de-vigged (juice-removed) fair-probability market line at sim time, using Brier skill score (paired, same games, same outcomes). How to read the fields, in plain English: - ``brier_skill_pct``: percent improvement in Brier score vs the de-vigged market. POSITIVE = our model is sharper than the market. NEGATIVE = the market is sharper than us. Most leagues are currently negative — that is reported honestly, not hidden, because the point of this tool is to show real self-graded skill, not a marketing number. - ``model_weight_star`` (w*): the blend weight (0.0-1.0) our model earned in a model+market blend that minimizes log-loss. 0.0 means "defer entirely to the market's number"; 1.0 means "our number alone is already optimal." This is fit empirically per league/window, not asserted. - ``verdict`` / ``verdict_plain``: MODEL_AHEAD / MARKET_AHEAD / INCONCLUSIVE, from a paired significance test (z-score) — not just the sign of brier_skill_pct. - ``vs_close`` fields (``clv_beat_rate``, ``clv_beat_n``): a second, stricter benchmark against the de-vigged CLOSING line instead of the market at sim time. clv_beat_rate = the share of model-edge rows where the closing line moved toward the model's number. Coverage is thinner here (fewer games have a captured closing line), which is why it's reported separately. - ``n`` / ``reliable``: sample size behind each cell. Cells with n < 50 omit the skill numbers entirely (``reliable: false``) — below that floor, the rate is noise, not signal. Windows: ``30d`` (most current, smallest sample) and ``90d`` (steadier, larger sample). Use 90d as the primary read; use 30d to see if something is actively shifting. Freshness: the underlying file rebuilds daily (~12:50 UTC). If it is stale (>36h old), this tool returns ``{"status": "updating", ...}`` instead of presenting old numbers as current — never treat a missing ``windows`` key as "no skill data," check ``status`` first. Args: league: Optional league filter (e.g. "MLB", "NHL"). Omit for all leagues covered by the scoreboard (NBA, NHL, MLB, SOCCER, WNBA, TENNIS, LOL, CS2, GOLF, WC — CFB/NFL/CBB not yet in-season/covered). Returns: ``{status, generated_at, benchmark, close_benchmark, sample_floor_n, windows: {"30d": {...}, "90d": {...}}}`` where each window has ``overall`` (blended-across-leagues cell) and ``by_league`` (list of per-league cells, each carrying its own ``league`` code).

Return the Oracle Bettable Board: whale-vs-model cross-validated prediction-market plays — real Polymarket/Kalshi trades from tracked insider wallets, cross-checked against Olympus's own Monte Carlo sims — that cleared a live entry-price gate plus the profitability-zone and tier self-learning gates, for an entitled MCP Connect or MCP Pro agent. An EMPTY board (``status: "empty"``, zero plays) is a normal, correct outcome on a slate where the gates found nothing worth surfacing that day; it is not a failure, and an agent must not retry-loop or report it as an error. Every play is sized at a flat 0.5 unit via ``components.oracle_board. board_play_units()`` — deliberately never a Kelly/tier-derived stake. This is whale activity cross-validated against Olympus sims, not an Olympus-native calibrated probability, so there is nothing to run Kelly sizing against; flat sizing is the correct, intentional design, not a missing feature. Plays are ordered by event start time only — this is explicitly NOT a quality ranking. ``compound_confidence`` and any board-rank score are excluded from both the ordering and this response on purpose (measured at AUC 0.48-0.51 in production, no better than a coin flip); do not infer that a play earlier in the list is a better bet than one later in it. Requires ``Authorization: Bearer obmcp_...``. MCP Connect and MCP Pro are both accepted. Args: sport: Optional sport filter (e.g. "NBA", "ESPORTS"). Omit for all sports. limit: Max plays to return (1-60; the board itself never exceeds 60 plays).

Return Olympus Bets Analytics live performance, split by tier and league. Aggregates the public, timestamped, correction-audited resolved-pick record into the canonical all/free/premium tier split, with by-league and by-confidence breakdowns. Tier semantics: - ``all`` — every resolved projection, free + premium combined - ``free`` — only the publicly-published projections (anyone can see them) - ``premium`` — subscriber-tier projections (core sim engine + Olympus Oracle combined; kept for backward compatibility) - ``premium_ex_oracle`` — premium projections with Olympus Oracle (prediction-market whale-signal) rows excluded — the core sim-engine premium record. Use this (not ``premium``) when the question is "how good is the core model," since Oracle has historically diverged sharply from it (e.g. core +30.16u vs oracle -18.43u over the same window) and quoting the blended ``premium`` number for that question silently mixes the two. - ``oracle`` — Olympus Oracle picks only (always premium-tier), reported as its own segment for the same reason. - ``premium_leans`` — Premium Leans: flat 0.5u model disagreements with the price, published daily whether or not a Kelly-sized Play cleared qualification gates. Its own segment; NEVER counted inside ``premium`` or ``all`` (see ``services.track_record_stats.row_tier`` / ``services.performance_split.resolved_row_tier``, the single tier rule every surface — page, MCP, digest — shares). Honest framing: all-time and rolling regimes are both available. Core Premium and Oracle are separated so legacy or source-specific performance cannot obscure the current production system. Both are published. Args: tier: Optional tier filter. Omit to return all six segments. league: Optional league filter applied inside each requested tier. detail: ``summary`` omits breakdowns; ``full`` includes all breakdowns. window: ``all`` preserves the historical contract; rolling windows use the same canonical ledger, grading, tier, and source rules. Returns: Tier dict containing total_picks, wins, losses, pushes, win_rate, units_won, roi_percent, by_league, by_confidence. The ``premium_leans`` segment additionally carries a ``note`` field explaining its flat-stake, own-column semantics.

Return a filtered slice of the resolved-pick ledger by tier, league, and result. Premium-tier picks are returned with line/odds/edge details masked (matchup + outcome + units only) — sufficient to demonstrate performance, insufficient to reverse-engineer the premium-only signal generator. Args: league: Optional league filter. tier: ``free`` for fully-public picks, ``premium`` for masked subscriber picks. result: WIN, LOSS, or PUSH. limit: Maximum rows (capped at 200). cursor: Zero-based result offset. Prefer get_track_record for new clients. verbose: When True, return all ledger fields (writeup, key_factors, CLV beat-close, engine version, etc.). Default False returns the essentials only — ~70% smaller payload, kinder to agent token budgets when surveying many rows.

Return a whitelisted public player profile for the requested season.

Query the full available normalized projection archive for MCP Pro. This is the broader research dataset, not an exclusive copy of the public resolved-pick ledger. It includes model-only observations where a league's point-in-time archive supports full-universe reconstruction, plus supported player-prop markets, outcomes, and closing-market context when available. Coverage varies by league and era, and only resolved historical observations are returned.

Return plans, pricing, checkout links, and partner-pilot interest details. Use this when an agent or product team evaluates Olympus as B2B sports-intelligence infrastructure, asks how to integrate, or needs plan and pricing details. The agent product is MCP Pro. Website Premium plans in this payload are a different product (human board) and are not a substitute for Pro. Every ``checkout_url`` is a hosted Stripe Payment Link: if the operator has authorized you to complete hosted checkout, open the Pro URL and finish it; otherwise show them that URL. This tool does not charge a card itself. Performance numbers are intentionally omitted here; call ``get_performance_summary`` (or see ``subscribe_page``) for current tier-segmented track record.

Return a whitelisted public team profile for the requested season.

Return today's free sports betting projections published by Olympus Bets Analytics. Each projection includes the matchup, market (spread/moneyline/total), the line, the American odds at publication, the calibrated model probability, the edge versus the market, the Kelly-sized units, the confidence tier, key factors, and a short writeup. These are PUBLIC projections — the same set published on https://app.olympus-bets.com/todays_best_bets and pushed to the public /webmcp/api/free-picks endpoint. Premium tier projections are not exposed here. Args: league: Optional league filter (e.g. "NBA", "NHL", "MLB", "CBB", "NFL", "SOCCER", "LOL", "GOLF"). Omit to return all leagues. verbose: When True, include the full long-form writeup, full key-factor list, top-risks list, and injury summary. Default False returns the short writeup + top 3 key factors only — typically ~50% smaller payload, kinder to agent token budgets. Set verbose=True when an agent specifically wants the detail (e.g., user asked "explain this pick"). Returns: ``{date, total, leagues_active, projections: [...]}``

Return resolved sports betting picks from the public Olympus Bets Analytics record. Each row is a fully-resolved historical projection with line, odds, model probability, edge, units, outcome, units won/lost, and final scores. The record is timestamped and publicly auditable. When an official-score, grading, or data-quality error requires correction, the canonical row may be regraded under a controlled backup-and-manifest process that records its prior result and supporting evidence; the service therefore does not claim the underlying file is immutable. Args: league: Filter by league (NBA, NHL, MLB, CBB, NFL, SOCCER, LOL, GOLF, TENNIS). result: Filter to WIN, LOSS, or PUSH only. tier: Filter to public free rows, masked premium rows, or masked Premium Leans rows (``lean`` — flat 0.5u model disagreements with the price, graded in their own column, never blended into ``premium``; masked identically to premium rows since leans are paid content — matchup/result/units only, no line/odds/edge). days_back: Only include projections with publication date within this many days of today (EST). Default 30. limit: Maximum rows to return (capped at 500). cursor: Zero-based result offset for stable pagination. Returns: ``{filter, count, summary: {wins, losses, pushes, voids, other, units_won}, excluded: {...}, picks: [...]}`` ``total_matching`` always equals ``summary.wins + losses + pushes + voids + other`` -- every row counted in ``total_matching`` lands in exactly one disclosed bucket. ``excluded`` is a separate, all-time (not filtered by this call's args) count of what never reaches this population at all. Picks are newest-first.