Sportily

Save a horse to the user's personal My Picks tracker so its result and ROI are tracked over time. Provide the Sportily raceId and horseId (from search_races / get_race_card / discover_* tools). The pick is auto-tagged with the source (claude or chatgpt) so the user can filter by who suggested it. Idempotent — saving the same runner twice will not duplicate it. Use only when the user explicitly asks to save, track, or add a pick.

add_to_my_picks

Analyse the user's My Picks track record: overall strike rate, stake-weighted ROI, beat-SP (closing-line value), max drawdown and win/lose streaks, plus a performance breakdown grouped by a chosen dimension (going, course, country, distance, class, type, or tag) and plain-language best/weakest insights. Use when the user wants to understand patterns in their picks — e.g. 'how do my picks do on soft going' or 'where am I losing money'.

analyze_my_picks

Backtest one of the authenticated user's saved Race Lens filters, or a draft filterAst, over historical completed races. Race Lens model predicates use the base price-agnostic race model, not SmartPicks or Value Finder tipster percentages. Capped at 60 days for MCP (see the returned `days` / `maxBacktestDays`). Note: get_saved_lenses_summary reports metrics over each lens's own original lookback (which can be longer than 60 days), so its numbers will not exactly match this tool's — compare lenses using results computed by this tool over the same `days`. Returns performance, sample selections, and concrete filterAst improvement ideas to try next.

backtest_race_lens

Use for horse-vs-horse comparison in a Sportily race. Returns focused evidence for selected runners; treat model percentages as directional analytics signals, not absolute truth.

compare_runners

Use when a user asks for broad horse-racing picks, good horses, value, SmartPicks, Value Finder, or Sportily analytics candidates. This is a candidate-discovery step, not a final verdict: it returns a shortlist you must then open. Call once per date/source/filter set in a user turn, then — if the user asked anything evaluative (worth backing, which is best, what is overrated) — immediately call get_race_card for the top 3 raceIds in the same turn and answer from those. Do not offer to open them and wait; do not rank the shortlist as a substitute for a verdict. Only return a bare list if the user explicitly asks for a list.

discover_racing_picks

Use when a horse-racing user names a horse and the Sportily horseId/raceId is not yet known. Finds runners by partial horse name, or lists runners for a date so the MCP client can match a horse. Once the runner is resolved, do not repeat this lookup for the same horse in the same user turn.

find_horses

Default tool for "what races are on?" — lists a full day's racecourses and race ids grouped by course, with today/tomorrow selection and an upcomingOnly filter. Use search_races instead when the user already named a specific course/date and you only need a quick flat list of matching race ids.

find_races

Fetch validated industry context for a Sportily race. Use at most once after get_race_card when externalResearchTargets are available and outside context would materially help. Returns status, source URLs, and extracted facts; cite the source when using verified external facts.

get_external_race_context

Use after finding a Sportily horse id when a horse-racing user asks for an opinion, profile, form view, or analysis of a named horse. Fetches career stats, upcoming races, and recent runs; open the race card before a race verdict.

get_horse_profile

Use when a raceId is known and the user asks to analyse a horse race, Sportily race, runner, racecard, or named horse in a specific race. Fetches the primary pre-race analytics payload with race, runner, model, market price, stats, past-performance, response-shape, and external-research guidance. Call once per raceId in a user turn unless the previous call failed. Target-race result fields are intentionally omitted so historical analysis can be run blind.

get_race_card