ai.shumi/mcp

Ask Shumi any crypto-market question in natural language. Shumi classifies the query, fetches the relevant data, and returns a synthesized answer. Use this when no specific typed tool fits, or for multi-part / comparative questions.

Daily snapshots of the tracked basket — composition and performance over time.

Detail for one category. view=info (trend breakdown), coins (member coins), sentiment.

Market cap, volume, open interest, funding rate and price for a coin AS OF a chosen point in the past. Returns a single snapshot, not a series: call it once per point you want to compare (e.g. amount=7 interval=d for a week ago, then again with no offset for now). Use this to answer "how has funding/open interest changed since…".

Bundled risk context for one or more coins: price, funding APR, deterministic funding_paying_side, funding_receiving_side, carry_if_long and carry_if_short fields, daily/weekly trend, sentiment stance, and BTC correlation. Funding applies to perpetual positions only; spot positions neither pay nor receive it. Relay the carry fields exactly rather than inferring direction from crowding. The best single tool for "should I be worried about X".

On-chain/social sentiment aggregates for a single coin.

Discrete funding-rate alert events (asset, trigger zone, funding at trigger, fired-at time).

Perpetual funding-rate positioning context, market-wide or for one symbol. APR is already in percent units (1.7 = 1.7%). Positive funding means longs pay and shorts receive; negative funding means shorts pay and longs receive. Funding is context only, not a standalone directional, timing, or entry signal.

Perpetual-futures signal engine. view=state (currently open signals), log (recent fires), history (one asset's past signals — needs asset).

Global market aggregates: BTC dominance, total market cap, total volume.

Tracked token-holder cohorts. view=watchlist (which token contracts are tracked), movements (recent holder-count changes for one contract). Answers "is the holder base growing or bleeding?".

The most recent regime crossing — when one trend cohort (UP/HODL/DOWN) overtook another market-wide. Answers "has the market flipped?". Returns an empty `crossings` array plus a message when no crossing has fired.

Overall market health: the UP/HODL/DOWN trend distribution and extreme movers. Set context for the full bundle.

Aggregate market sentiment. view=market/latest/summary (overall), narratives, categories, slopes/entity-slopes (what is trending), health (pipeline status).

Sentiment and momentum for a single named narrative.

Pair-trading and delta-neutral funding-arbitrage intelligence. mode=suggestions (pair ideas), delta-neutral (funding arb), history (backtest), signal (state for a specific pair — needs token_a & token_b).

Bulk live prices, optionally with 4h/24h/7d baseline overlays. Omit symbols for the full tracked set.

Market regime signals. view=active (current positions), signals (all), confidence (scores). Provide symbol to get that symbol's regime history instead.

Price, daily/weekly trend and perp funding for one real-world asset (stock, ETF, commodity, index, FX). Look up by ticker (AAPL, GOLD) or by namespaced id (xyz:AAPL). Funding belongs to the PERPETUAL CONTRACT, not the underlying — `funding.apr` is the annualized rate in percent, `funding.rate` is the raw per-hour fraction. Do not use the crypto coin tools for these.

Synthesized verdict for a coin, combining trend, funding, sentiment and regime.

Signal validation envelope for an asset: Sharpe ratio, win rate, sample size, reliability tier.

Highlights mined from tracked video/podcast transcripts. view=highlights (extracted claims with the coins, sectors and macro tags they mention), sources (which channels are tracked). Answers "what are people actually saying about X?".

Upcoming token unlocks ACROSS THE MARKET when no single coin is named: each row has date, days until, coin, tokens, USD value, % of market cap, recipients and insider (team + investor) share, with its source. Defaults: next 7 days, top 500 coins, at least 0.5 % of market cap, ranked by % of market cap. Only coins DropsTab tracks; each recipient's next release only. Answers "biggest unlocks this week?", "big VC unlocks this month?". For one coin use get_unlocks.

Scheduled token unlocks for ONE coin: the next unlock (date, days until, tokens, USD value, % of market cap, recipients such as team/investors/ecosystem, cliff or monthly tranche), what releases over 30 and 90 days, % unlocked and locked, the final unlock date, the last unlock, and how price moved in the 7 days before and after its last 3-6 unlocks (history, not a forecast). Sources: DropsTab cross-checked with DefiLlama, each figure with its source link and fetch time; `state` says `upcoming`, `fully_unlocked`, `no_unlock_data` or `not_covered` (then `publishedSchedule` may carry the project's own vesting terms). Answers "when is ARB's next unlock?", "who gets the OP unlock?", "how did SUI trade around its unlocks?".

Tracked wallets. view=watchlist (which wallets are tracked), movements (recent balance changes for one wallet address). Answers "what did this wallet do recently?".

List tracked crypto categories by their exact names (e.g. "Meme", "Layer-2"), the names scan_coins expects.

List the currently active market narratives (e.g. "AI coins", "DeFi summer").

List the tradable real-world assets — stocks and ETFs (AAPL, NVDA, SPY), metals and commodities (GOLD, SILVER, BRENT), stock indices (SP500, JP225) and FX. These trade as perps on Hyperliquid builder DEXes and are NOT crypto tokens; the crypto tools will not find them. Use this to answer "which stocks/commodities can I look at?".

Look up a single coin and its core metrics (price, trend, metadata) by symbol, name, CoinGecko/internal id, or on-chain contract address. Read the coin's CURRENT trend from `currentTrend` ({ trend: UP|DOWN|HODL, since, days, asOf, incompleteDayExcluded }; `currentTrendWeekly` when present is the weekly one), not from the last row of `trends` — `trends` is the history of past trend runs. Quote it as "<trend> since <since> (<days> days, as of <asOf>)". Only when `currentTrend` is absent, fall back to the last `trends` row.

Fuzzily resolve a symbol, name, or contract to canonical coin candidates. Use this first when the user input is ambiguous, before calling other tools.

Filter the tracked universe by trend direction, category and market-cap band, and sort the result. For movers questions ("what's pumping", "top gainers/losers today", "biggest movers") use sort_by="change24h" ("change7d" for the week) — sort_order="desc" for gainers, "asc" for losers — and quote each row's change from the row itself (`change_24h_pct` / `change_7d_pct`, or `change24h` / `change7d`). With the default marketCap sort rows are plain coin names; a change sort may return `{ rows: [...] }` with a coverage summary instead of a bare list. Pass ONE category per call; the server ignores case and punctuation ("meme", "Layer 2" and "DeFi" all work) and rejects an unknown name with close matches (list_categories has the names). There is no exchange filter yet: `exchange` is rejected with an explanation, because the only venue data behind this scan is spot listings. Do not present a scan as "coins on <exchange>". An empty result means no coin matched all the filters.

Trend scanner. state=fresh (newly started), stale (longest running), aligned (multi-timeframe agreement), extreme (biggest moves), historical.

Search the web for crypto information, or get a direct answer. Backed by Shumi's web-search tool.