live_long_short_ratio is Free; market_price and market_orderbook are Free, market_recent_candles is Starter and market_historical_oi is Pro; book_summary is Starter and the other book tools, including the trigger-order tools, are Pro; the four Sharpe tools are Starter; the archetype tools are split between Starter and Pro. Where a tier cell below links to the matrix, check the matrix for that tool’s current tier. Check the tier matrix for the endpoint a tool wraps rather than inferring from its name.
The connector registers 115 tools; builder_heatmap is withheld while its hourly rollup backfills, so clients see 114.
Renamed in v0.17.5 and v0.17.6. Eleven tools have new, descriptive names. The old names still work as aliases (same inputs, plan and output) but are no longer listed by the connector, so update saved prompts and agents to the new names:
book_* tools answering a question each, rather than handing you a raw book to parse. v0.11.0 added builder analytics: eleven builder_* tools covering the leaderboard, per-builder profiles, attributed traders and fills, cohort composition, retention, audience overlap, user lifecycle, journey economics and order intent, plus the inverse per-wallet view.
Builder analytics (v0.11.0)
Every builder tool (all twelve, including
trader_builders) carries the attribution caveat in its response (dataNotes): trigger-order (stop/TP) fills attribute at ledger level only until the parser upgrade completes, so attributed detail slightly undercounts exact ledger revenue.
Discovery & taxonomy
Pulse — trader & cohort analytics
Pulse — single trader
In clients that support MCP Apps interactive views,
pulse_trader_profile, pulse_trader_performance and pulse_trader_position_summary render the Coinversa wallet card (90-day position stats, PnL curve, open positions); other clients receive the data as text.
Pulse — cohorts
Sharpe ratios (v0.17.4)
All four Sharpe tools are Starter. An annualized realized-PnL Sharpe — √365 × mean daily realized PnL ÷ its sample standard deviation, days without fills counted as zero, today excluded. It is a proxy, not a portfolio Sharpe: not divided by equity, and fees, funding and unrealized PnL are left out. It rewards consistency over size, so read it next torealizedPnl. See Sharpe ratios.
Sharpe filters on cohort tools.
live_cohort_bias, live_positioning_by_entry, pulse_cohort_positions, pulse_cohort_trades and the five pulse_cohort_recent_* tools take optional minRecentSharpe (30-day score, refreshed hourly) and minTotalSharpe (full-history score, refreshed daily). Both are inclusive minimums, combined with AND, applied before limits and aggregation, and use each wallet’s current score. Unlike the leaderboard they have no activity floor: one winning day scores about 3.49, so a lower threshold can admit one-day wallets. Each tool keeps its own tier.
Pulse — position lifecycles
Every position reconstructed open→close from on-chain fills, with MAE/MFE. See Position lifecycles.Pulse — execution quality (MAE/MFE) · Pro
Pulse — trader archetypes
Pulse — market structure
Pulse — refreshed cohorts (30-day rolling tier)
Cohorts labelled by last-30-day tier (pnl_tier_recent / size_tier_recent) — who’s printing right now, not all-time. All five are Pro.
Live — real-time market + risk
live_long_short_ratio is Free. The snapshot tools (what is true right now) are Starter. History and the chart-layer tools are Pro.
In MCP Apps clients,
live_cohort_bias renders a long-vs-short-by-tier view and live_liquidation_heatmap renders a liquidation price ladder; the text result is unchanged.
Market — direct Hyperliquid passthrough
Interactive chart (v0.17.0)
Released 12 Oct 2026 in connector v0.17.0; the sections below cover the chart work from v0.14.0 to v0.17.0. The interactive views appear only in clients that support MCP Apps; text-only clients get the same tool results.
v0.14.0: price charts and bounded candle history
Usemarket_recent_candles for a price chart; use market_price for a quick mark/index/funding/OI lookup. Candles require Starter; the chart’s Coinversa layers (entry profile, average entries, Apex vs Crowd history, biggest trades) are Pro, and on a Starter key the chart marks them with an upgrade link instead of failing. An MCP Apps client can render candlesticks, dollar volume, moving averages (EMA 21, SMA 20/50), VWAP, Bollinger Bands, RSI, RSI zone markers, MACD and Stochastic. Text-only clients retain the candle result; indicators selects the initial chart layers without changing those rows.
The limit bounds the source history too. With the current fallback aggregation,
5m and 15m use the latest 500 one-minute bars; 4h and 1d use the latest 500 hourly bars. You may therefore receive fewer than 500 aggregated candles. This is recent history, with no older-page cursor.
Market search uses list_markets: search matches the symbol case-insensitively before pagination, and dex narrows the venue. For example, search for gold, then use the returned xyz:GOLD ticker. Inactive listings are hidden unless includeInactive=true; the connector’s search and paging controls do not add parameters to the underlying API route.
v0.15.0: explore the chart locally
The chart adds a searchable market picker, remembered selections, synchronized indicator panes, candle readouts, keyboard inspection, pan/zoom, and controls to return to the latest available bars. Search typing, indicator settings and range changes use loaded data. Returning to a cached market/interval reuses its candles; Refresh requests candles explicitly and keeps the previous chart visible if the request fails. Liquidation levels and Apex vs The Crowd positioning load on request and refresh separately from candles. They describe current snapshots; each layer identifies its retrieval time and any available source measurement time. Expansion and sending an analysis prompt are available only when the host supports them; otherwise the prompt remains selectable for copying. Standard candle VWAP usessum(HLC3 × base quantity) / sum(base quantity), where HLC3 = (high + low + close) / 3. Intraday intervals reset at the UTC day; daily candles anchor to the first loaded bar. It is available only when the source supplies base quantity for the loaded candles and there is positive volume. If those weights are missing, the chart reports VWAP as unavailable instead of treating USD notional as base quantity. The volume histogram remains in dollars.
Chart layers (v0.16.0)
The chart now draws what only Coinversa has: where open positions got in, who holds them, and how the two flagship cohorts have moved. Two new tools carry the data, and both are useful without the chart:live_cohort_bias_history gains tiers (up to 8, either vocabulary — not together with tier), format: "layer" (the API’s chart envelope with a −1..1 netBias, per-series hoursPresent and provenance, returned unpaged), tierType: "pnl_recent" and hours up to 2160 (90 days). A format: "layer" answer can be up to 2,160 points per series — pass maxResponseTokens or a shorter hours when reading it as text; a shortened answer says so in _responseLimit, and partial data should never be read as complete. format: "layer" requires tier or tiers. tierType: "pnl_recent" snapshots begin 2026-05-14, so no earlier hours are returned.
All three are snapshots or hourly series with explicit caveats — current open positions are never history, whale-trade cohorts are the wallets’ current labels and the aggressor is unknown, and hourly history may contain reconstructed hours with today’s labels, so it is not suitable for backtesting. The chart layer endpoints page has the fields and the reasoning.
In the chart. Ask for a chart — “show me the BTC 4h chart” — and market_recent_candles opens the interactive view in clients that support MCP Apps. Under Layers, the Coinversa layers sit beside the technical indicators:
With no
indicators passed, the view opens with RSI, the entry profile, average entry lines and liquidation levels. indicators accepts ema21, sma20, sma50, vwap, bollinger, rsi, macd, stochastic, rsiMarkers and the Coinversa layers liquidations, positioning, entryProfile, averageEntries, cohortHistory and whaleTrades to choose the initial layers. A PnL grouping selector switches the entry layers and history between all-time and 30-day PnL cohorts; the Apex vs Crowd snapshot and biggest-trade markers always use all-time groups.
Each layer loads once per market and is cached in the view. Refreshing candles does not refresh layers — Refresh layers does that explicitly — and there is no polling. Every layer shows its source and as-of time, and Data & sources keeps the full coverage notes. What Claude sees about the chart is a concise summary per layer, never raw fills or wallet addresses.
Image sharing (v0.17.0)
Share, beside the price, opens a fullscreen composer. Pick a card — Chart + inspection (the chart as you left it, with the selected candle or trade) or Cohort positioning (Apex and Crowd long/short shares, open notional, wallet counts, average entries and leverage) — and a format: X landscape (1600 × 1200) or Instagram portrait (1080 × 1350). Then Share image hands the PNG and an editable caption to your device’s share sheet where the device supports it, and Save image downloads it. Images are rendered locally in the view: no upload service, no account integration, and nothing is posted automatically. Each card carries its snapshot time and the footer current positions, not a forecast and Market data from Hyperliquid, not investment advice. Complete coverage notes stay in Data & sources.Entity resolution — owner-level intelligence · Pro new in 0.9
Hyperliquid traders run named sub-accounts that trade independently — per-wallet views undercount every multi-account trader. These tools resolve wallets to their owner entity and aggregate across the whole family.Exchange aggregates new in 0.9
Plans & limits new in 0.9
When the plan does not include a tool
Every gated tool’s description ends with its plan (Plan: Starter ($29/month) or higher.). If a key’s plan does not include the tool, the result is an ordinary tool result, not a tool error (isError is false), shaped like this:
partialAnswer is present only when the key’s plan includes a tool that answers part of the question. Examples: the current price and long/short ratio for the same coin, the Starter cohort-bias snapshot for a Pro positioning layer, or book_summary for a Pro book tool. Agents should state the plan and price, share upgradeUrl once, use the partial answer (and say which tool it came from), and not retry the gated tool. Revoked or invalid keys are still reported as errors. On REST, the same denial is the 403 described in Authentication.
HIP-4 — outcome contracts
L4 order book (v0.12.0)
Four tools over the order-level book — every resting order with the wallet behind it — each shaped around a question rather than returning a book to parse yourself.book_summary is Starter; the other three are Pro.
The wallet addresses join to the trader tools.
book_whales returns addresses that go straight into pulse_trader_profile or pulse_trader_performance — so “who is holding up this bid, and are they any good” is one follow-up call, not a research project.
Bounds: book_stop_map takes within_pct 1–10 (default 10), book_whales takes limit 1–50 (default 20), book_levels takes depth 1–100 (default 50). All three are integers — a fraction is rejected before a request is made.
Snapshot-derived, latest-only. The gateway re-exports the node’s full L4 book on a fixed cadence and the rollups are recomputed per coin, so there is no history — one row per coin, replaced each time its block height advances. Every response carries
as_of_height (the L1 block the answer is true at) and age_s; check age_s before citing a number. A rollup older than 1,200 seconds (20 minutes) returns 503 rather than a stale book. For the aggregated L2 view with no tier gate, use market_orderbook.data_coverage reports a book dataset with coins covered, the latest height and its freshness.l4Book websocket feed.
TP/SL trigger orders (v0.17.4)
Both tools are Pro.book_stop_map buckets stop and take-profit size by distance from mid; these two tools return the individual orders behind it, with the wallet on each. See TP/SL trigger orders.
They read their own exchange-wide snapshot — native perps, builder dexes and HIP-4, independent of the L4 sweep — refreshed roughly every minute, latest only. A snapshot missing or older than 1,200 seconds (20 minutes) returns
503. As with the book tools the coin is case-sensitive and not normalized, but an unknown coin returns an empty list rather than a 404, so an empty answer proves nothing. side is the order’s B/A (buy/sell), not the position’s direction.
Paging: limit 1–500 (default 100), offset up to 100,000. For any page after the first, pass snapshot_height = the first page’s as_of_height; a 409 means the snapshot changed, so restart at offset 0. Each result adds count, hasMore, nextOffset and a note with the value to send next.
Connector-side parameters
Several tools take shaping parameters that the connector applies after it fetches the endpoint — the routes themselves take none of them, and you will not find them in the API reference or the OpenAPI spec. Calling the endpoint directly returns the full payload.
Parenthesised numbers are the default and the maximum. Every capped response carries
totalCount, truncated and a note naming the next offset, so an agent can page without guessing.
Around twenty other tools also take limit and offset, and those are passed straight through to the route, which pages for itself. The distinction only matters if you are calling the REST API directly: for the tools above, the parameter names in the Connector-side column do not exist on the endpoint.
Every tool accepts a
useToonFormat flag (default true) that returns data in compact TOON format — about 50% fewer tokens than JSON for tabular data. Agents typically want this on; set to false if you need standard JSON.Every tool also accepts maxResponseTokens (500–30,000; complete array items are removed until the answer fits, disclosed in _responseLimit) and responseTruncation (head or tail, for unpaged arrays). Both are applied by the connector.