crypto-quant-signal-mcp

The call intelligence layer for AI trading agents — composite quant calls across crypto and TradFi perps, cross-venue arbitrage detection, and regime-aware market classification via MCP.

Live Track Record — 90%+ directional accuracy across 900+ trade calls. Public, no login required.

Why AlgoVault

Most MCP trading servers give you raw data — prices, order books, candles. Your agent still has to figure out what to do with it.

AlgoVault is different. We give your agent one answer: a directional verdict with a confidence score, built from a multi-factor composite scoring engine tuned on production quant systems. Every call is tracked, every outcome is measured, and the full track record is public from day one.

What makes this not just another indicator wrapper:

• Composite scoring, not single-indicator noise. Multiple orthogonal signals — momentum oscillators, trend structure, derivatives positioning, volume dynamics, open interest flow — fused into a single weighted verdict. The weights are calibrated from live market outcome data, not textbook defaults.

• Regime-aware call generation. Calls are filtered through a market regime classifier before emission. The engine knows when to issue calls and when to stay silent — a trend-following setup in a ranging market gets suppressed, not broadcast.

• Cross-venue intelligence. Funding rates from Hyperliquid, Binance, and Bybit are normalized and compared in real-time. Nobody else does cross-venue derivatives analysis via MCP.

• Published track record with every release. Every call is recorded with outcome prices at multiple horizons. Win rate, profit factor, and expected value are computed continuously. No cherry-picking, no survivorship bias.

• Adaptive scoring. Indicator weights are retuned monthly from outcome data. The engine learns what works and adjusts — the call you get today is better than the one from last month.

• Crypto + TradFi coverage. 290+ assets — standard crypto perps, TradFi perpetuals (stocks, indices, commodities, FX via Hyperliquid's xyz dex), and liquidity-filtered meme coins. Assets are classified into quality tiers with institutional-grade call filtering.

Try It in 30 Seconds

No code. No API key. No install.

Step 1. Open Claude → Settings → Integrations → Add custom connector

Step 2. Enter the name and URL:

Step 3. Ask Claude anything:

"Get me a trade signal for ETH on the 4h timeframe"

That's it. Your Claude now has a quant analyst built in.

Tools

get_trade_signal

Returns a composite BUY / SELL / HOLD verdict with confidence score for any supported asset — crypto perps, TradFi perpetuals (stocks, indices, commodities, FX), and liquidity-filtered meme coins on Hyperliquid.

Under the hood: a multi-factor scoring engine evaluates momentum, trend structure, derivatives sentiment, open interest dynamics, and volume conviction. Scores pass through regime-aware filters and adaptive post-processing gates — including funding flow analysis, volatility regime detection, and trend persistence decay — before a final verdict is emitted.

Only high-conviction calls are generated. The engine is designed to stay silent when the edge is unclear.

Parameters:

• coin (string, required): Asset symbol — e.g. "ETH", "BTC", "SOL", "GOLD", "TSLA", or any of 290+ supported assets

• timeframe (string, default "15m"): "1m", "3m", "5m", "15m", "30m", "1h", "2h", "4h", "8h", "12h", "1d"

• includeReasoning (boolean, default true): Human-readable explanation of the call logic

Output includes: call direction, confidence score (0–100), all computed indicator values, detected market regime, reasoning narrative, and _algovault metadata for downstream tool composability.

scan_funding_arb

Scans cross-venue funding rate differentials across Hyperliquid, Binance, and Bybit. Normalizes hourly vs 8-hour rate conventions, computes basis-point spreads, and ranks opportunities by composite score (spread magnitude, time urgency, and funding conviction from 24h history).

This is the only MCP server that provides cross-venue funding arbitrage intelligence — long one exchange, short another, capture the spread.

Parameters:

• minSpreadBps (number, default 5): Minimum spread in basis points to include

• limit (number, default 10): Maximum results returned

Output includes: per-opportunity venue rates, optimal long/short direction, annualized spread percentage, and next funding timestamps.

get_market_regime

Classifies the current market environment into one of four regimes: TRENDING_UP, TRENDING_DOWN, RANGING, or VOLATILE.

Uses a multi-dimensional classification approach combining directional strength measurement with ADX slope analysis (detecting trend strengthening vs exhaustion), volume-weighted pivot detection, ATR-adaptive funding thresholds, and cross-venue funding sentiment divergence. The regime classification directly informs how get_trade_signal filters its output — agents can also use it independently for strategy selection and position sizing.

Parameters:

• coin (string, required): Asset symbol

• timeframe (string, default "4h"): Candle timeframe for analysis

Output includes: regime label, confidence score, underlying metrics (trend strength, volatility interpretation, price structure), cross-venue funding sentiment, and a plain-English strategy suggestion.

Performance Tracking

Every call is tracked from emission to outcome. No exceptions.

What we measure:

• Outcome prices at timeframe-appropriate evaluation windows

• PFE Win Rate — did price move in the call direction at any point during the evaluation window

• Expected Value — probability-weighted average return per call

• Profit Factor — gross wins divided by gross losses

• Peak Favorable Excursion (PFE) and Maximum Adverse Excursion (MAE)

• Running statistics per asset, timeframe, and quality tier

HOLD calls are free — when the engine says "don't trade," you don't pay. Only BUY and SELL verdicts are charged via x402 or count against subscription quotas. This aligns our incentives: you only pay when we see a tradeable opportunity.

• HOLD Rate: Percentage of scans where the engine declines to issue a trade call. A high HOLD rate (currently ~84%) means the engine is selective — it only calls BUY/SELL when conditions align across multiple indicators.

Infrastructure:

• Remote mode: PostgreSQL with automated outcome backfill

• Local mode: SQLite at ~/.crypto-quant-signal/performance.db

• Only high-confidence BUY/SELL calls are tracked — HOLD is excluded

On-Chain Verification

Every call is hashed (keccak256) at creation time and anchored on Base L2 via daily Merkle batches. This makes the track record tamper-proof — we cannot edit past calls.

• Contract: 0x6485...0f81 (Base L2)

• Verify any call: https://api.algovault.com/api/verify-signal?signalId=<ID>

• View all batches: https://api.algovault.com/api/merkle-batches

• Visual verification: algovault.com/verify

Pricing

* HOLD verdicts (engine says "don't trade") are always free across all tiers — no x402 charge, no quota deduction.

x402 micropayments: AI agents pay per HTTP call with USDC on Base — no signup, no API key, no billing. The payment receipt is the credential. See x402.org.

Subscriptions: Sign up at api.algovault.com/signup. Starter ($9.99/mo) unlocks all assets and timeframes. API key delivered instantly after checkout.

For Developers

Remote endpoint (recommended)

https://api.algovault.com/mcp

Streamable HTTP transport. Compatible with any MCP client — Claude, Cursor, Cline, custom agents.

Local install via npx

npx -y crypto-quant-signal-mcp

Claude Desktop / Cursor config

{
  "mcpServers": {
    "crypto-quant-signal": {
      "command": "npx",
      "args": ["-y", "crypto-quant-signal-mcp"],
      "env": { "TRANSPORT": "stdio" }
    }
  }
}

npm install

npm install crypto-quant-signal-mcp

Self-hosting

git clone https://github.com/AlgoVaultLabs/crypto-quant-signal-mcp
cd crypto-quant-signal-mcp
cp .env.example .env  # Edit with your values
npm ci && npm run build
docker compose up -d

Architecture

Agent / Claude / Cursor
  │
  ▼
api.algovault.com/mcp (Streamable HTTP)
  │
  ├─ x402 payment verification (USDC on Base)
  ├─ API key / subscription check
  ├─ Free tier fallback
  │
  ▼
MCP Server (Express + @modelcontextprotocol/sdk)
  │
  ├─ Composite Scoring Engine
  │    ├─ Multi-factor indicator fusion
  │    ├─ Regime-aware signal filtering
  │    └─ Adaptive post-processing gates
  │
  ├─ Asset Classification Engine
  │    ├─ 4-tier quality system (Blue Chip → Major Alt → TradFi → Meme)
  │    └─ Liquidity filter for meme/micro assets
  │
  ├─ Exchange Adapter Layer
  │    └─ Hyperliquid (standard + xyz TradFi perps) · Binance, Bybit (funding data)
  │
  ├─ Performance Tracker
  │    └─ PostgreSQL (remote) / SQLite (local)
  │
  └─ Hyperliquid Public API (free, no auth)

Exchange adapter pattern: All exchange interactions go through the ExchangeAdapter interface — supporting both standard crypto perps and xyz TradFi perps on Hyperliquid, with Binance and Bybit for cross-venue funding comparison.

Suite Composability

Every tool output includes an _algovault metadata block declaring version and compatible downstream tools:

Schemas are designed for composability. All tools share consistent timestamp, coin, and _algovault fields — downstream tools accept these objects directly as input.

Privacy

Local mode: Zero telemetry. No data sent to AlgoVault servers. Call history stored on your machine only.

Remote mode: Request metadata logged for analytics (IP hashed, never stored raw). See privacy policy.

License

MIT

Built by AlgoVault Labs — call intelligence for the autonomous trading era.

Landing page · API endpoint