MCP server · autonomous market researcher · streamable HTTP
Spawn your real-time market researcher.
Turn any AI client into an autonomous market analyst and hypothesis researcher for BTC, ETH, SOL and ZEC. Connected to our MCP server, your agent continuously scans market memory, formulates competing price-level and session/event hypotheses, replays historical touches with fees against a baseline, and messages you only when it discovers a statistically unusual setup. Start with research_market_memory, run_candle_research_watch or scan_candle_memory_radar.
All MCP tools are free for everyone while we study demand. There is no daily quota or set end date. No key or card needed.
https://aipricepatterns.com/api/mcpclaude mcp add --transport http aipp https://aipricepatterns.com/api/mcpInteractive MCP Protocol Topology · Click Any Node to Execute Live
Explore & invoke MCP tools directly on the graph
scan_candle_memory_radar
Scans BTC, ETH, SOL and ZEC across timeframes and ranks windows by how tightly 20 historical twins agree on magnitude.
AIPP Matrix Sentinel · Real-Time Market Researcher
Scanning 9 windows (BTC/ETH/SOL × 15m/1h/4h) · 20 historical twins per window
run_candle_research_watch- RADARBTCUSDT 1h (q=5, f=10): 20 historical twins agree on terminal magnitude (38% as wide as all-history).
- BASELINEWalk-forward get_candle_track_record verified: analogue band beats unconditional baseline (+21% Winkler lift, 120 anchors).
- FILTER7 noisier timeframe windows filtered out automatically before bothering the user.
Awaken this researcher in Claude, Cursor, ChatGPT or Telegram
Connect https://aipricepatterns.com/api/mcp or launch 24/7 in Telegram:
You are my AI Price Patterns Market Research Entity (AIPP Matrix Sentinel · Real-Time Market Researcher). Read get_api_guide, topic watch. Every 15m, Run run_candle_research_watch across BTCUSDT, ETHUSDT and SOLUSDT (15m, 1h, 4h) with watchId 'sentinel-matrix-v1'. When status is ANOMALY_READY and shouldMessageUser is true, also check investigate_candle_event_hypothesis for UTC session or event-window clustering and message me in this chat with the market, anchorUTC, tested hypotheses (including rejected ones), counts, and the /candles link. Otherwise stay quiet. Never name a side or a size, and end every note with NO_TRADE.
run_candle_research_watch + investigate_candle_event_hypothesisBrowser Lab
Try the research before connecting an agent
Enter your own price level in the browser. Inspect reach counts, target or invalidation order and the dated examples behind them.
Backtest hypotheses with RLXBT Server (Docker)
AIPP gives your AI agent market memory. To test execution rules with realistic commissions, slippage, Monte Carlo proof, and RL, deploy the headless RLXBT Docker container. A shared Pro license key is available free for agents.
tool: get_rlxbt_backtester_trialOptional account key
Connect anonymously with the URL above. An account key is optional during the free access experiment.
No card. You can skip this and use the server anonymously without a daily quota.
Access
- Anonymous — free, no daily quota
- Account key — optional, same free access
- Experiment — end date not set
The guide and the evidence board never count.
Connect
claude mcp add --transport http aipp https://aipricepatterns.com/api/mcp
No account key required during the free access experiment.
Leave the agent on watch
Connect once. Then ask the agent to repeat the research on its own schedule and write to you in that same chat only when history looks unusual. Claude, Cursor, ChatGPT, Grok and any other MCP client can do this. The server does not send the alert, and it does not name a buy or a sell. A client that only answers when you type will wait until you ask.
You are connected to AI Price Patterns (BTCUSDT, ETHUSDT, SOLUSDT, ZECUSDT). Read get_api_guide, topic watch (and topic market-query for custom causal queries). On your schedule, run that standing watch: call run_candle_research_watch with a persistent watchId (or radar, pulse of the top row, track record for the same window, and replay_candle_hypothesis when the pulse names a target and an invalidation) and check investigate_candle_event_hypothesis or research_market_memory (including relative_return / observe.level_touch) when an anomaly appears. Message me in this chat only when that guide's bar is crossed. Otherwise stay quiet. Never name a side or a size. End a note with NO_TRADE and the /candles link.
The bar the agent follows is in get_api_guide, topic watch. A note names the market, the counts and the chart link, and it ends in NO_TRADE.
How your MCP agent reads the Constellation Radar & syncs with /candles
Instead of averaging all 20 historical twins into one flat coin-flip band, the server projects the 20 closest matches into a 2D cosine-similarity map around NOW (0,0) and partitions them into 3 structural clusters (0 = Bullish Expansion, 1 = Range / Chop, 2 = Pullback / Fade). Your connected agent can inspect multi-timeframe cluster alignment, verify cluster-scoped brackets with riskUsd position-size arithmetic, and push setups directly onto your open /candles chart.
5m·15m·1h·4hget_candle_constellation_mtf
Checks all 4 timeframes for BTCUSDT, ETHUSDT, SOLUSDT or ZECUSDT. Reports whether ≥3/4 timeframes share the same dominant cluster (aligned vs conflict) and whether the 3-bar nowDrift vector points into that cluster.
Cluster 0/1/2get_candle_live_pulse
Returns constellation (candle-memory-constellation/v1): each cluster’s isolated P10 / P50 / P90 outcome band, member timestamps, and dominant feature driver (Last Bar + Volume, 96b Context, 2-Bar Flow).
clusterId + riskUsdcheck_candle_order
Pass the trader’s levelPrice, invalidationPrice, targetPrice plus clusterId: 0 | 1 | 2 and riskUsd. Returns clusterScope first-touch counts and riskArithmetic (positionSizeUnits, notional USD, R:R).
Browser Bridgepush_candle_workspace_action
Read what is open on /candles via get_active_candle_workspace, then push a cluster cone focus, plot a 3-price bracket on the chart, and pin an R-multiple trade idea with an agentNote.
You are connected to AI Price Patterns MCP (https://aipricepatterns.com/api/mcp). Read get_api_guide with topic "constellation". Follow this 4-step Constellation Radar workflow: 1. Call get_candle_constellation_mtf for BTCUSDT, ETHUSDT, SOLUSDT and ZECUSDT (5m, 15m, 1h, 4h). Look for markets where confluence.status is "aligned" (>=3/4 timeframes share the same dominant cluster: 0 = Bullish Expansion or 2 = Pullback / Fade) and the 3-bar nowDrift vector points into that dominant cluster. 2. Call get_candle_live_pulse on the aligned timeframe. Quote brief verbatim (including the monospace chart and NO_TRADE) and inspect constellation.clusters[dominantClusterId] for its isolated P10/P50/P90 outcome band and similarityTag. 3. When I name an entry (levelPrice), invalidation (invalidationPrice) and target (targetPrice) — or when you inspect the bracket currently open on my chart via get_active_candle_workspace — call check_candle_order with clusterId set to the active Constellation cluster (0, 1 or 2) and riskUsd (e.g. 100). Report clusterScope first-touch counts (target first vs invalidation first) and riskArithmetic (positionSizeUnits, positionNotionalUsd, rewardRiskRatio). 4. Use push_candle_workspace_action to focus the matching cluster cone on my open /candles browser tab, plot the 3-price bracket on my chart, and pin the setup with an agentNote so I can review or route it to execution. Keep all historical evidence strictly NO_TRADE.
Complete Closed Loop · Observe → Verify → Execute Safely
Connect the full 3-MCP stack: AIPP + RLXBT + ManusPay
Wire three focused MCP servers into one agent so market memory, statistical rule proof, and budgeted on-chain execution stay strictly separated.
aipricepatterns.com/api/mcp
Detects crowd extremes (get_crowd_positioning) and US-session momentum (get_intraday_momentum), applies 2D Constellation risk gates (C2_VETO_SQUEEZE, Cluster 1 Chop Veto), and syncs setups to /candles.
127.0.0.1:8142/api/mcp (RLXBT)
Runs event-driven backtests, Walk-Forward Analysis, and Monte Carlo survival with fees and slippage. Kills overfit ideas with NO EDGE FOUND before any capital is risked.
manuspay.xyz (Local MCP Wallet)
Executes Solana swaps & transfers inside pre-set per-trade and daily USD caps with pre-flight simulation and request_id idempotency. Anything larger pauses for your Touch ID.
{
"mcpServers": {
"aipp": {
"type": "http",
"url": "https://aipricepatterns.com/api/mcp"
},
"rlxbt": {
"type": "http",
"url": "http://127.0.0.1:8142/api/mcp",
"headers": {
"Authorization": "Bearer ${RLXBT_API_KEY}"
}
}
}
}Run the 3-MCP closed-loop workflow (AIPP -> RLXBT -> ManusPay): 1. [AIPP] Call get_crowd_positioning, get_intraday_momentum, and get_candle_constellation_mtf. Check whether a live trigger is active and passes the Constellation risk filter (C2_VETO_SQUEEZE = false, not inside dominant Cluster 1 Chop). Push the bracket to /candles with push_candle_workspace_action. 2. [RLXBT] Before sizing any trade, verify the rule and fees in local RLXBT MCP (127.0.0.1:8142/api/mcp) using Walk-Forward Analysis and Monte Carlo. If RLXBT reports NO EDGE FOUND or OVERFIT, stop immediately. 3. [ManusPay] Only when both AIPP's structural gate and RLXBT's proof pass, propose or execute the bounded Solana swap inside ManusPay's per-operation and daily budget with pre-flight simulation and a unique request_id (pausing for Touch ID above the cap).
What your agent can ask
New: 2D Constellation MTF · Live /candles Workspace Sync · market-query/v0 · sweep
get_candle_constellation_mtfNEW · MTF RADARMulti-timeframe (5m · 15m · 1h · 4h) 2D Memory Constellation alignment for BTC, ETH, SOL or ZEC: partitions the 20 closest twins into 3 structural clusters (0 = Bullish Expansion, 1 = Range / Chop, 2 = Pullback / Fade), reports each cluster's P10/P50/P90 band and similarityTag, computes the 3-bar NOW drift vector, and flags aligned vs conflict.
get_active_candle_workspaceNEW · LIVE SYNCRead what the trader currently has open in their /candles browser tab: active symbol, timeframe, selected Constellation cluster (0/1/2), 3-price bracket (levelPrice, invalidationPrice, targetPrice, riskUsd), pinned R-multiple trade ideas, and browser neuralProfile.
push_candle_workspace_actionNEW · LIVE SYNCPush a live visual update directly to the trader's open /candles browser tab: switch symbol/interval, focus a Constellation cluster cone (0/1/2), plot a 3-price bracket with riskUsd on the chart, and pin the idea to the top bar with an agentNote.
research_market_memoryNEW · v0Typed causal market-query/v0 engine: search events (excursion, compression, volume_spike, clock, percentile, relative_return), sequences (A -> B -> C), or exact similarity; observe same-asset, cross-market (observe.on), spread reversion/continuation (observe.versus), or prev_day/prev_week level_touch order (high_first, low_first, same_bar, any_touched) with Wilson/Newcombe CIs against baseline.
reuse_market_researchNEW · SWEEPSave immutable public research templates, reuse saved query_id / report_id / complete anchor sets across markets & date ranges, or run an atomic Cartesian sweep (up to 12 cells) across BTC, ETH, SOL, ZEC and timeframes.
get_forward_validationLIVE · AUDITIndependently verify and follow our live pre-registered systematic strategies (Flagship HN9_ALTS9_40H & sleeves) without exposing proprietary formulas: OOS benchmark corridor, live net bps / hit rate / drawdown, active open signals, and timestamped closed trades verifiable on Binance.
compare_relative_returnNEW · ZEC & HYPECompare one crypto against another across BTCUSDT, ETHUSDT, SOLUSDT, ZECUSDT, and HYPEUSDT over fixed non-overlapping UTC windows (6h, 12h, etc.): signed relative return (difference/ratio) or absolute magnitude comparison (abs_difference: abs(base) > abs(versus)), win share with 95% CI, spread quantiles, by-year split, and volatility regime split.
get_api_guideGUIDEBuilt-in agent manual & schema discovery. Read topic 'constellation' for the 2D Radar & browser sync, 'watch' for standing alerts, or 'candles' for the full start-here guide.
run_candle_research_watchOne-call autonomous researcher patrol. Scans the BTC/ETH/SOL/ZEC matrix, checks track-record lift, generates and replays competing upside P90 and downside P10 hypotheses, checks UTC session/era concentration, and deduplicates alerts via watchId.
scan_candle_memory_radarFirst call. No arguments. Ranks BTC, ETH, SOL and ZEC by how tightly past twins agree on the size of the next move. Not a direction.
Try: “Run research_market_memory on SOLUSDT 1h where SOL outperformed BTC by 3%+ over 12 bars, and observe the SOL-BTC spread over the next 24 bars.”
Or: “Check ZECUSDT 1h prev_day high/low level_touch order at 00:00 UTC versus all-history baseline, and sweep across BTC, ETH, SOL and ZEC.”
Market Query v0 · Ready-to-Use Instructions
Copyable research recipes for your agent
Switch between natural-language chat prompts or raw JSON for research_market_memory and reuse_market_research. Strict causality, no lookahead, explicit NO_TRADE framing.
1. MTF 2D Constellation Radar Alignment
Check 5m · 15m · 1h · 4h cluster dominance, P10/P50/P90 bands & 3-bar NOW drift vector
2. Cluster-Scoped Bracket & Risk $ Sizer
Verify first-touch barrier order inside Cluster 0 (Bullish Expansion) with $100 risk sizing
3. Push Setup & Cluster Cone to /candles Tab
Highlight a Constellation cluster and plot + pin a 3-price bracket on the open browser chart
4. Cross-Market Divergence & Spread
SOLUSDT outperforming BTCUSDT by ≥3% over 12h → observe 24h spread reversion/continuation
5. Causal Level Touch Matrix
Count prev_day High/Low first-touch order (high_first, low_first, same_bar, any_touched) on BTCUSDT
6. 4-Coin Atomic Sweep (incl. ZEC)
reuse_market_research action: sweep across BTC, ETH, SOL and ZEC on one pinned snapshot
Check the quality of the range estimate
Range estimates are compared with an all-history baseline. Results depend on the configuration; a historical level-touch count does not establish a profitable strategy.
See the evidence board →