# midasflow-mcp-quickstart MCP server

Agent-native crypto market-data over MCP+REST: order flow, whales, liquidations, calibrated scores

## Links
- Registry page: https://www.getdrio.com/mcp/io-github-midasflowai-lab-midasflow-mcp-quickstart
- Repository: https://github.com/midasflowai-lab/midasflow-mcp-quickstart
- Website: https://midasflow.ai/docs/api

## Install
- Endpoint: https://mcp.midasflow.ai/mcp
- Auth: Not captured

## Setup notes
- Remote endpoint: https://mcp.midasflow.ai/mcp

## Tools
- get_signals (Get Flow signals (live feed / elite crown)) - Pull recent MidasFlow SignalEvent records (newest first). quality='normal' (default) = the bucketed signal log across source families; quality='elite' = the rare highest-conviction multi-model-consensus crown feed (premium+, lagged anti-front-run). Tier-gated + moat-scrubbed server-side. Market DATA, not advice. Routes: normal→/v1/signals/pull, elite→/v1/signals/elite. Endpoint: https://mcp.midasflow.ai/mcp
- score_symbol (Score symbol (calibrated P(TP1) + plan + safety)) - CROWN data product: calibrated P(first TP1 before SL) band + coarse trade plan (TP ladder %, SL %, weights) for ONE symbol, FOLDED with a pre-trade feed-safety check (is the feed live/real-volume/fresh, no phantom ticks). Coverage is present-or-null — most symbols return p_tp1_band=null (valid, NOT an error). Feed the band into calc_ev; never read it as buy/sell. Routes: /v1/score/{symbol} + folds /v1/symbol/check as `safety`. Endpoint: https://mcp.midasflow.ai/mcp
- get_candles (Get OHLCV candles (price history)) - Recent OHLCV candle bars for a symbol at a chosen timeframe — the raw price/volume history other tools are derived from. Inspect trend, range, volatility, volume profile, or feed your own indicators. Market DATA, not advice. A symbol outside the candle store returns an empty bars list (normal, not an error). Routes: /v1/candles/{symbol}. Endpoint: https://mcp.midasflow.ai/mcp
- get_orderbook (Get order-book snapshot (walls / imbalance + S/R)) - Live order-book snapshot for a symbol — bid/ask walls, book imbalance, spread — FOLDED with support/resistance levels (classic TA fused with cross-exchange book walls + Fibonacci) so you see resting liquidity AND the level map in one call. Cross-exchange aggregate, derived levels only. Market DATA, not advice; thin/uncovered symbols return null fields (normal). Routes: /v1/orderbook/{symbol} + folds /v1/sr as `sr`. Endpoint: https://mcp.midasflow.ai/mcp
- get_flow (Get order-flow (cvd / sweeps / cross / vpin) [grouped]) - Per-symbol order-flow microstructure, grouped by `kind`. kind='vpin'/'all' (default)=full snapshot (CVD trend+divergence, VPIN toxicity, sweep state, buy/whale aggression, /v1/flow); 'cvd'=CVD series (/v1/cvd); 'sweeps'=stop-sweep events (/v1/sweeps); 'cross'=cross-exchange flow origin (/v1/cross_flow). Realtime market CONTEXT, NOT advice / a win-rate. Empty/unknown kind → a menu of kinds. VPIN is ONE contested signal — pair with get_accuracy before treating it as edge. Endpoint: https://mcp.midasflow.ai/mcp
- get_heatmap (Get liquidation heatmap (price-magnet zones)) - Liquidation-heatmap matrix for a symbol — price levels where leveraged positions cluster and are likely force-liquidated (price-magnet zones / liquidity pools). Anticipate where a move may accelerate or stall. Market DATA, not advice; a symbol with no clustering returns an empty matrix (normal, not an error). Routes: /v1/heatmap/{symbol}. Endpoint: https://mcp.midasflow.ai/mcp
- get_context (Get context (regime / phase / derivatives / funding / squeeze / intel) [grouped]) - Contextual market reads, grouped by `kind`. kind='regime'=market-regime labels (/v1/regime, market-wide, no symbol needed); 'phase'=move-lifecycle / entry-timing for a symbol (/v1/phase, premium+); 'derivatives'=normalized cross-exchange funding/OI/basis summary (/v1/derivatives); 'funding'=PER-VENUE funding+OI (/v1/funding); 'squeeze'=liquidation-cascade proximity (/v1/squeeze); 'intel'=per-symbol aggregated signal-quality roll-up (/v1/intel). Market DATA / context, NOT advice and NOT a win-rate. Empty/unknown kind → a menu of kinds. Endpoint: https://mcp.midasflow.ai/mcp
- get_whales (Get whales / liquidations (large-print intel) [grouped]) - Large-player / forced-flow intel, grouped by `kind`. kind='whales' (default)=recent large ('whale') prints for a symbol, banded size+side (/v1/whales); kind='liquidations'=multi-exchange liquidation prints over a rolling window, long-liq vs short-liq notional bands per venue (/v1/liquidations; empty symbol = market-wide top movers by liq notional). Raw notional is banded. Market DATA, NOT a signal/advice. Empty/unknown kind → a menu of kinds. Endpoint: https://mcp.midasflow.ai/mcp
- get_market (Get market board (movers / overview / anomalies) [grouped]) - Whole-market reads, grouped by `kind`. kind='movers' (default)=top movers ranked (/v1/movers); 'overview'=whole-market roll-up — breadth / direction / activity (/v1/market); 'anomalies'=per-symbol or cross-market anomaly board (volume_spike/abnormal_spread/phantom_tick) with severity bands (/v1/anomalies). Cheap top-level CONTEXT to find what's MOVING before drilling into one symbol. Market DATA, NOT advice. Empty/unknown kind → a menu of kinds. Endpoint: https://mcp.midasflow.ai/mcp
- get_accuracy (Get accuracy / outcomes (proof of edge) [grouped]) - Realized track-record (proof of edge, forward-only, Wilson-bounded), grouped by `kind`. kind='accuracy' (default)=source×boost first-TP win-rate + boost lift in the canonical TP1 frame (/v1/accuracy); kind='outcomes'=realized aggregate track-record over a window from resolved signal_outcomes, model-free (/v1/outcomes). Win-rate frame = P(price hit TP1 before SL), NOT realized PnL. Model internals never exposed. Empty/unknown kind → a menu of kinds. Endpoint: https://mcp.midasflow.ai/mcp
- account (Account (tier / quota / balance / usage / tiers) [grouped]) - Your mf_ key's account view, grouped by `kind`. kind='account' (default)=tier/product, weighted daily quota, used/remaining today, billing, prepaid balance (/v1/account); kind='usage'=today's count + cap AND the endpoint_weights map (exact per-call cost units, /v1/usage); kind='tiers'=the Flow data-product tier ladder, delivered-now vs roadmap (/v1/tiers). Read account/usage to self-throttle by your remaining quota. Empty/unknown kind → a menu of kinds. Endpoint: https://mcp.midasflow.ai/mcp
- backtest ([LAB] Backtest a setup (historical P(TP1 before SL) + EV)) - LAB / RESEARCH tool — replay an arbitrary {symbol, entry, tp, sl} setup over MidasFlow's 1m candle history and get its historical P(TP1-before-SL) + EV, in the canonical first-touch TP1 frame (same frame as get_accuracy). Historical market DATA, NOT a prediction or advice. Low-sample setups return a directional band (normal, not an error). NOTE: live results are gated behind ff:backtest_live — until that flips, EVERY tier gets a coming-soon envelope (no live numbers). Routes: POST /v1/backtest. Endpoint: https://mcp.midasflow.ai/mcp
- analyze ([LAB] Analyze symbol (AI-Analyst report — heaviest)) - LAB / RESEARCH tool — the full MidasFlow AI-Analyst report for a symbol: a synthesized narrative fusing flow, derivatives, levels, regime and signal context into a human-readable read RIGHT NOW. This is the HEAVIEST call (composes many sub-feeds) so it costs MORE units and requires a TOP tier; call it sparingly, AFTER cheaper context tools. Market DATA / AI-generated analytics, NOT financial advice. Routes: /v1/analyze/{symbol}. Endpoint: https://mcp.midasflow.ai/mcp
- calc_ev ([LAB] Calculate expected value (local analytics, no network)) - LAB tool — pure local expected-value calculator over a win-rate and average win/loss percentages. No network, no advice, 0 units. Feed a get_accuracy bucket win-rate (or a score_symbol band) to reason about expectation per trade. Endpoint: https://mcp.midasflow.ai/mcp

## Resources
- midasflow://catalog - MidasFlow tool catalog (что для чего) Structured index of every MidasFlow MCP tool: what it's for, when to use it, its kinds, cost weight, min tier, and the TRADING (core 11) vs LAB (3) split. Read this first to pick the right tool. MIME type: application/json
- midasflow://guide - MidasFlow getting-started guide Short connect→discover→detect→confirm→score→size getting-started guide for an agent new to the MidasFlow Flow MCP. MIME type: text/markdown

## Prompts
Not captured

## Metadata
- Owner: io.github.midasflowai-lab
- Version: 1.0.0
- Runtime: Streamable Http
- Transports: HTTP
- License: Not captured
- Language: Not captured
- Stars: Not captured
- Updated: Jun 29, 2026
- Source: https://registry.modelcontextprotocol.io