N0brains MCP Server

[CONTEXT] Nearest historical market-state analogs to right now: k-NN over the cross-asset state (SPX/NDX momentum, VIX level + term structure, DXY, yield curve) with what SPX/NDX/BTC actually did over the following 1d/5d (median, quartiles, hit-rate) per analog and in aggregate. k = 3-25 (default 12), episode-separated. Same data as REST /analogs. Conditioning context, NOT a prediction.

[START HERE — market overview] One-call morning brief: market regime (risk appetite), liquidity read, high-impact events next 72h, the engine's actionable reads (has_trade_signal + reads[]), and cross-asset trade-plan ranking — compact projections of get_market_regime / get_liquidity_map / get_economic_calendar / get_actionable_signals / rank_trades, assembled server-side. Optional coin arg scopes the actionable reads to that coin and adds it to the ranked set (ranking stays setup_score-sorted). Drill into any block with the underlying tool. Descriptive + engine verdicts; uncalibrated blocks labeled; not financial advice.

[DRILL-DOWN] Latest TradFi market open prices for BTC/ETH/SOL across sessions. Source: watchers.market_opening_watcher.get_latest_opens(). Same data as REST /market-opens.

[CONTEXT — market overview] Market-wide risk-appetite read: risk-on / risk-off / squeeze from a blend of the macro composite, cross-sectional breadth, funding regime and vol. Answers 'do conditions favor risk right now?'. Mirrors REST /regime. Descriptive, uncalibrated, not financial advice. One-call morning brief incl. this block: get_market_brief.

[CONTEXT] Current macro bias (regime, BTC/ETH bias + conviction, calendar risks). Mirrors REST /macro current snapshot. Honesty overlay applied: fields marked uncalibrated, insufficient-data flags surfaced.

[CONTEXT] Active signals whose direction AGREES with the current macro bias (conviction ≥ 0.6). Uses the same rule the internal pipeline uses to boost confidence x1.12 (vs CONFLICTS, which dampens x0.88). macro and macro_pulse signal types are excluded (they ARE the macro). Optional asset filter.

[DRILL-DOWN] Mindshare leaderboard: each asset's share of crypto attention across n0brains' sources over the window, ranked, with velocity (rising / falling / emerging). The edge is a coin's attention ACCELERATING before price moves. Directional proxy over n0brains sources, NOT a market-wide social-firehose absolute. Same data as REST /mindshare.

[DRILL-DOWN] One coin's mindshare: its attention share %, rank, and velocity vs the prior window (rising/falling/stable/emerging). Same data as REST /mindshare/{coin}.

[RAW FEED — detail] Fetch a single signal by ID with full enrichment (historical_edge, paired_inverse, latency, priced_in fields). Returns 404 semantics via tool error if signal not found.

[RAW FEED — detail+context] Composite call: signal + same-asset S/R levels + active macro bias. Saves 2-3 round trips. Returns a dict (not a typed model — the composite shape varies).

[MONITORING] Polling alternative to the /stream WebSocket. Walks the signal stream in emission order (timestamp, id) from a cursor: pass `since_timestamp` (unix epoch seconds) and, to resume exactly, the `since_id` from the previous call's `cursor_id` (or parse `next_cursor` = 'timestamp:id'). Returns every signal emitted after the cursor — including ones that have since expired (see `status`) — up to `limit` (1-100, default 100), oldest first, with `has_more`. Keep calling while `has_more` is true. Free tier sees the standard delay; nothing is skipped, it arrives once the delay elapses. `server_timestamp` is still returned but is NOT a safe cursor (it moved past signals the page could not hold) — use next_cursor / cursor_timestamp + cursor_id.

[RAW FEED — engine inputs, NOT trade calls] List active n0brains signals with optional filters. Filters: asset (e.g. 'ETH'), signal_type (whale|sentiment|listing|regulatory|macro|macro_pulse|liquidation|funding|hack|price|other), direction (bullish|bearish|neutral), urgency (high|medium|low), min_confidence, min_score, limit (1-100, default 20), offset. Each signal includes historical_edge, paired_inverse, signal_latency_secs, priced_in_*, calibration_inverted_in_cell. CONFIDENCE CONTRACT: confidence = calibrated empirical win-probability estimate (binned per signal_type), NOT raw model output; when confidence is null, confidence_v2 + outcome_record (2026-09-02) = this asset x type x direction x context's measured 24h outcome record with a day-clustered interval, peer rank and `distinguishable` — quote it only with its n/days/ci95; confidence_suppressed_reason says why; confidence_status is one of calibrated|floor_demoted_at_emit|suppressed_anti_predictive|demoted_anti_predictive_type. Transform emitters (whale_position leaderboard fade) carry observed_direction/observed_behavior/model_transform/predicted_direction so the raw observation is never lost. Most rows carry action_hint=ignore — engine inputs, not calls; read historical_edge (cell win_rate) before echoing any direction. For tradeable output use get_actionable_signals.

[START HERE — coin snapshot] Unified whole-system snapshot for one coin: current price, per-coin market-state consensus (proven-voter directional read), nearest support/resistance levels, liq-map target/invalidation, and the shared macro regime (deterministic FRED composite anchor + LLM read + any divergence). One call instead of stitching get_macro + get_market_state + get_levels. Drill down only if needed: get_positioning (who is crowded), get_indicators (momentum+Fib), get_liquidation_map (magnets), get_options (vol). Pro tier. Measured + AI data, not advice.

[DRILL-DOWN — prose brief] LLM-written 'state of <coin>' in markdown: joins the headline consensus, macro composite, levels, technical indicators and flow context into one readable analysis you can quote to a user directly (the narrative layer over get_state; the structured payload rides along in `data`). Cached 15 min server-side. Mirrors REST /state/{coin}/brief. Pro. AI-generated synthesis, not advice.

[START HERE — 'vet my trade'] Ask n0brains First: graded pre-trade conditions assessment for a proposed trade. Give asset + side (long/short); optionally entry, stop, target, leverage, horizon_hours (default 24). Returns grade A..F with flags (positioning crowding, scheduled event risk inside the horizon, liquidation distance vs realized daily volatility, stop inside noise range, proven-edge conflicts, late entry), supporting factors, and falsifiers to watch. Grades are logged and resolved at horizon; cross-grade performance stays withheld until the deployed weekly truth gate clears. Pair with get_positioning (who is crowded) + get_event_outlook (scheduled risk inside the horizon) for the full vet. Analytical, not advice.

[STEP 2 — plan one coin] Assembled trade plan for one coin: direction, entry, strongest target, stop, risk/reward, sizing hint, options context (put/call + skew), and warnings (max-pain timing against the trade, entry near a liq cluster). Mirrors GET /plan/{coin}. Analytical, not advice.

[START HERE — 'find me a trade'] Signals the production trade-gate itself marked actionable (action_hint=trade_signal — the engine's per-(type,direction) proven verdict), age ≤ max_age_min. Skips anti-predictive cells. Also returns swing_outlooks: labeled days-scale reads from cells proven at 7-30d horizons, each with its proven horizon and a suggested hold — NOT intraday trades. Pass min_score / min_confidence only if you want additional numeric bars on top of the engine verdict. When the result is empty, the `context` block points to rank_trades / get_trade_plan — a signal-gate miss does not mean no setup exists (positioning/levels setups aren't signal-driven). ALWAYS check `has_trade_signal` (true only when a real intraday trade cleared the gate) and render `reads` — a single array that is NEVER empty when any read exists: it holds the gate-passed trade signals, or, when none cleared, the strongest context read + swing outlooks, each tagged `kind` and `actionable`. `signals` stays strictly gate-passed; items in `reads` with actionable=false are NOT trades. Next: grade any candidate with check_trade; rank_trades when nothing cleared the gate.

[JOURNAL] Amend an OPEN journal trade by trade_id: move your stop or target, fix size_usd / leverage / thesis. A stop MOVE changes only the current stop (what the watchdog and close-time touch scan use); realized R stays measured against your INITIAL stop, so trailing to breakeven can't inflate R. To fix a genuine fat-finger in the original entry or stop, also pass correct_entry=true — that resets the R basis (disclosed in the response). asset/side can't be amended — void and re-log for that. Pro.

[RECEIPTS] Cells from cell_stats.json with inverse_flagged=true. These are (signal_type × direction × regime) buckets where the empirical win-rate is below the inverse_thresholds floor with sufficient sample. Signals in these cells get calibration_inverted_in_cell=true and have confidence nulled in customer-facing serialization.

[RECEIPTS] Per-non-crypto-asset-class forward-return scoreboard (asset_class = stock | index | metal | commodity). Measured on that class's own rows + baseline (stock excess vs SP500; index/metal/commodity absolute). Intel-only: the tradeable badge is informational, non-crypto is not auto-traded yet. status=accruing until a (type,direction) reaches the min sample. Same data as REST /proof?asset_class=. For the crypto board use get_performance or REST /proof. Not financial advice.

[META] The asset universe check_trade / get_trade_plan / get_levels can price: Hyperliquid perp coins + tokenized HIP-3 stocks/metals/indices. Call once instead of discovering support by error. Contract: count === len(assets); `degraded: true` means the list is INCOMPLETE/STALE — do not treat it as the universe, do not cache it, retry later; no `degraded` key means healthy. Mirrors REST /check/assets (shared cache).

[JOURNAL] Close a journal trade by trade_id (from log_trade or get_journal). Optionally exit_price (defaults to live price) and note (exit reasoning). n0brains resolves the outcome from real candles over the held window: return %, R multiple vs your initial stop, MAE/MFE (worst drawdown / best unrealized gain while open), and whether your stop or target level actually traded. HISTORICAL backfill: pass closed_at (epoch seconds of the real exit) + explicit exit_price. Pro.

[DRILL-DOWN] Return-correlation + beta of a coin to BTC and ETH over a 7d window of 15m log returns, plus its most/least correlated peers. Descriptive statistic (correlation is not causation). Same data as REST /correlation/{coin}.

[CONTEXT] Cross-asset flows: crypto rotation, crypto-vs-tradfi OI split, institutional posture (ETF flow / COT / 13F, descriptive). Answers 'where are funds going and is the market buying something other than crypto?'. Mirrors REST /flows. ETF flow is proven non-predictive. Not advice.

[RECEIPTS — experimental] Emergent edge discovery: corroboration class-combinations mined from the shadow ledger vs realized forward returns, ranked by measured edge (honesty-gated, both-halves). Surfaces patterns nobody hand-coded. status=accruing until the ledger fills (~60-90d). Candidate, not advice.

[TIMING] Scheduled macro + earnings calendar — the 'knows WHEN' feed. Upcoming high-impact US macro releases (CPI, NFP, FOMC, PPI, GDP) and tracked single-name earnings (NVDA, TSLA, MSFT, +) with consensus/previous, and actual + surprise once printed. Args: days_back (0-90, default 7), days_ahead (0-60, default 14), event_class ('macro'|'earnings', optional). Same data as REST /calendar. Context for timing/regime, not a direction call.

[TIMING] Upcoming scheduled macro releases + earnings joined with each event's MEASURED historical reaction distribution (event-study library, grouped by surprise sign): 'CPI prints Thursday — the last N hot prints moved SPX/BTC X%'. history=null until a cell accrues (the library earns its conditionals, it never manufactures them). Same data as REST /event-outlook. Not a direction call.

[META] Liveness + lightweight pipeline stats: uptime, signals in last 1h, current macro regime, classifier backlog, local-LLM classify breaker state (mode/open/consecutive fails). Mirrors REST GET /health with extra context. Pro-gated (per tools/call rule) — use REST /health for unauthenticated liveness.

[RECEIPTS — your own] Your past check_trade assessments WITH resolved outcomes: each row is the trade as you submitted it, the grade it got, and (once the horizon passed) the side-adjusted result with stop-touch honored. This is your personal calibration on the CHECKS you asked for (the journal covers trades you actually took). Free tier also gets free_checks_remaining_today. Mirrors REST /checks/history (last 50). Analytical, not advice.

[DRILL-DOWN] Technical indicators for a coin (e.g. 'BTC', 'ETH', 'SOL', 'XRP'): RSI(14), MACD, SMA/EMA (20/50/200 + 200-week), Stochastic, and FIBONACCI retracement levels (90-day swing). Returns daily + weekly timeframes plus a plain-language read. Same data as REST /indicators/{coin}. Use for momentum + Fib confluence with get_levels.

[JOURNAL] Read your private trade journal. status=open|closed|void|all (default all shows open+closed; void is hidden unless asked), limit for history (default 20). Open trades include live unrealized PnL/R and a warning if your stop level has traded since entry. Closed trades include resolved outcomes (ret %, R, MAE/MFE). stats block = personal calibration: win rate and realized R per n0brains entry grade — where your entries were actually good. Pro.

[DRILL-DOWN] Support/resistance levels for a coin (e.g. 'BTC', 'ETH', 'SOL'). Reads from levels_engine + Hyperliquid mids. Levels are detected on THREE timeframes — 15m (3 days back), 4h (90 days) and 1d (2 years) — and every level carries its `timeframe`, `distance_pct` and zone band. `nearest_resistance`/`nearest_support` are the immediate intraday levels (within ~5% of spot) and are often fractions of a percent away; for swing structure read `structure.resistance` / `structure.support`, which are the 4h/1d levels out to 35% from spot, strongest first. A level merged across tiers lists the confirming timeframes in `confirmed_by`. Same data as REST /levels/{coin}.

[DRILL-DOWN] Liquidation map for a coin (e.g. 'BTC', 'ETH'), binned into price clusters — the same feed that powers positioning's liq_magnet and market_state's target/invalidation. Shows long/short imbalance per zone (long_usd vs short_usd per bucket), nearest dense cluster below and above price, and top zones by notional. PROVENANCE VARIES BY COIN — always read the returned `observed` / `modeled` / `method` fields before describing the data. BTC, ETH and HIP-3 tokenized stocks/metals/indices have a DEX book, so their maps are OBSERVED per-position liquidation prices (Hyperliquid + GMX). Coins with no DEX book (XRP, SOL, DOGE, most alts) return a MODELED estimate built from aggregate CEX open interest and calibrated leverage tiers — real zones, but an estimate, and its long/short totals are symmetric by construction. Same data as REST /liqmap/{coin}.

[CONTEXT] Net cross-asset liquidity map: Fed net liquidity, stablecoin dry-powder, total perp OI, liquidation pressure, net taker flow. Answers 'where is liquidity?'. Mirrors REST /liquidity. Descriptive, not advice.

[JOURNAL] Log a REAL trade entry into your private n0brains journal the moment it fills. Give asset + side (long/short); optionally entry (defaults to live price), stop, target, size_usd, leverage, thesis (why you took it). n0brains snapshots full entry conditions automatically (grade, flags, positioning, regime — an internal check_trade) so nothing needs hand-transcribing. HISTORICAL backfill: pass opened_at (epoch seconds of the real fill) + explicit entry; the entry grade is then taken from YOUR check_trade nearest the fill (±6h, same asset+side) — the read you actually got at the time, never re-graded on today's tape; no matched check = ungraded. Close with close_trade; read back with get_journal. Returns trade_id + the entry assessment. Journal is private to your account. Pro.

[DRILL-DOWN] Long/short positioning for one coin from REAL data, mode picked by the asset's primary source: DEX price-point buckets (Hyperliquid+GMX, BTC/ETH-style), CFTC COT (metals/oil/indices), or exchange long/short ratios (alts). Returns latest buckets {price, long_usd, short_usd}, totals + long_pct + ls_ratio, the accumulated trend over `days` (1-90, default 7), and funding + OI-by-venue context. Complements get_positioning (the 8-leg synthesis) with the raw who-is-long-where view. Mirrors REST /charts/long-short/{coin}. Analytical, not advice.

[FORENSICS] Manipulation-risk analysis for one asset: composite manipulation score, coordinated-pump probability, fake-engagement risk, liquidation-cascade detection. Richer than the per-signal manipulation_score field — this is the full standalone read. Mirrors REST /manipulation/{coin}. Analytical, not advice.

[FORENSICS] Narrative heatmap for one asset: signal momentum, velocity, decay, manipulation probability and directional conviction across 1h / 4h / 24h windows — is the story building or dying? Complements get_mindshare_coin (attention share) with time-structure. Mirrors REST /narrative/{coin}. Analytical, not advice.

[DRILL-DOWN] Options analytics for a coin (BTC or ETH): ATM implied vol, skew (put-call IV proxy — the fear gauge), IV term structure, put/call OI ratio, and max-pain, from public Deribit data. Positive skew = downside hedging/fear; term_structure slope > 0 = contango. Descriptive positioning, not prediction. Same data as REST /options/{coin}.

[RECEIPTS] Backtest performance over last N days (1-365, default 30). Same data as REST /performance. Note: no asset filter — performance is aggregated across all assets. Performance is the live forward-return record by signal type. DENOMINATOR CONTRACT — every block (overall, by_signal_type, by_coin, by_source) carries total/wins/partials/losses/expired/settled/win_rate. `total` counts signals emitted, settled or NOT: most expire unsettled, so wins/total is not a win rate and ranks lanes backwards. `win` = full take-profit only; a partial is not a loss. Rank on win_rate (wins/settled) or avg_pnl. `win_rate` is null when settled=0 — that is 'not yet measurable', NOT zero. For the gated per-type verdict with confidence intervals and the tradeable badge, use get_proof instead.

[READ FIRST] The routing guide for every n0brains tool: which tool answers which intent (find a trade / vet a trade / coin snapshot / market brief / monitoring) and how to interpret the honesty fields (action_hint, historical_edge, n_eff, calibration). Call this once if you are unsure which tool to use — it replaces trial-and-error over the 40-tool catalog. Static text, no market data, free tier.

[DRILL-DOWN — who is crowded] Positioning thesis for one coin: who is crowded and which way. Combines funding-rate crowding (30d z-score), taker CVD buy/sell dominance (2h + 24h), open-interest-vs-price divergence (new longs / short-covering / new shorts / capitulation), options put-call + skew + max-pain TREND (BTC/ETH/XRP/SOL), nearest liquidation magnets above/below with notional, and whale stance (fade-corrected) into a single net positioning bias in [-1,1] with plain-English reasoning per component, per-line data freshness, and a coverage grade (full/broad/partial/thin — how many of the 8 legs actually contributed; full means every leg, `missing` names any that did not, so a thin-coverage score cannot read like a full one). Use INSTEAD of manually combining get_liquidation_map + get_options + funding. Mirrors REST /positioning/{coin}. Conditioning context, not financial advice.

[TRUTH ANCHOR] THE canonical current price (live exchange mid) for a coin — the single source of truth every other n0brains tool's spot/current_price should agree with. Returns {coin, price, source, age_secs, ts}. Use this to sanity-check any analytic payload: if a tool's spot disagrees materially with this, that tool's price is stale and its read should be discounted. Free tier. Public data, not financial advice.

[TRUTH ANCHOR] Batch canonical prices for several coins in one call. coins = comma-separated symbols, e.g. 'BTC,ETH,SOL' (max 50). Returns {prices:{SYM:price}, missing:[...], source, age_secs}.

[RECEIPTS] The full public forward-return proof board — richer than get_performance: per-signal-type measured post-signal performance with the proven-gate `tradeable` badges, plus the SWING boards. Args: asset_class (stock|index|metal|commodity|fx — same as get_asset_class_proof) OR horizon ('7d'|'14d'|'30d' for the swing-horizon boards, measurement-only, never badged tradeable). Omit both for the default 24h crypto board. Same data as REST /proof and n0brains.com/proof. Measured, not advice.

[STEP 2 — pick the coin] Cross-asset ranking: assembled trade plans for the given coins sorted by setup_score (best first) — answers 'which coin is the better trade right now?'. coins = comma-separated (default BTC,ETH,SOL). Mirrors GET /rank. Analytical, not advice.

[CONTEXT] Altseason/rotation read: is capital rotating INTO alts (altseason) or back to BTC (risk-off)? rotation_score in [-1,1] from relative-strength breadth + correlation trend. Breadth is a PROXY, not true BTC dominance. Uncalibrated heuristic. Same data as REST /rotation.

[DRILL-DOWN] Aggregate sentiment for a coin: net directional lean (confidence-weighted, recency-decayed), chatter volume + velocity (is it accelerating?), and contributing sources, over 24h. Coverage is CURATED high-edge authors — what the tracked smart-money voices lean, NOT mass social volume. Same data as REST /sentiment/{coin}.

[DRILL-DOWN — history rhymes] Semantic similarity search across the signal corpus: give a coin and/or a free-text query (q), get the k most similar past signals ranked by embedding cosine similarity — 'have we seen this setup before and what did it look like'. k = 1-20 (default 5). Provide at least one of coin / q. Mirrors REST /signals/similar. Pro. Analytical, not advice.

[FORENSICS] Trust / scam-risk screen for a token: resolves the ticker to a contract via DexScreener, then checks GoPlus Security for honeypot behavior, mint function, high taxes, and insider concentration. Run this BEFORE taking any alt-coin signal seriously — a bullish read on a honeypot is worthless. Mirrors REST /trust/{coin}. Analytical data only, not financial advice.

[META] Your own API usage: total calls, per-day series and top endpoints over period '7d' or '30d'. Use it to budget calls — free tier check_trade is 3/day (get_check_history shows the remaining count). Mirrors REST /usage. Private to your account.

[JOURNAL] Void a mis-logged journal trade by trade_id (wrong asset, duplicate, fat-finger) with an optional reason. Soft-delete: the trade is removed from your stats and the default journal view but retained and recoverable (get_journal status='void' lists voided trades). Voiding a CLOSED trade removes its outcome from your calibration — disclosed in the response. Use this for entries that never should have existed; use close_trade for real trades that ended. Pro.