io.github.AlgoVaultFi/crypto-quant-signal-mcp

Quick start (30 seconds)

No code. No API key. No install. The server speaks Streamable HTTP at https://api.algovault.com/mcp — any Model Context Protocol client connects directly.

1. Add the connector. In Claude → Settings → Integrations → Add custom connector:

2. Ask for a call. In plain language:

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

Your Claude now has a quant analyst built in. Prefer local? Run npx -y crypto-quant-signal-mcp.

What one call returns

One API call. One verdict. Not 8 raw indicators. get_trade_call returns a directional BUY / SELL / HOLD with a confidence score and the detected market regime — a composite verdict, not a data dump for your agent to interpret.

Under the hood, a self-tuning model fuses momentum, trend structure, derivatives positioning, open interest, and volume into one weighted call. Weights are calibrated from live trade outcomes, not textbook defaults. Regime filters suppress low-edge setups, so the engine stays silent unless the signal is clear.

{
  "call": "BUY",
  "confidence": 78,
  "price": 84250.50,
  "indicators": {
    "funding_rate": 0.0001,
    "funding_24h_avg": 0.00008,
    "funding_state": "NORMAL",
    "oi_change_pct": 2.4,
    "volume_24h": 2381602633,
    "trend_persistence": "HIGH",
    "breakout_pending": "INACTIVE"
  },
  "regime": "TRENDING_UP",
  "reasoning": "Trending regime, upward bias. Funding pressure mild. Volatility neither expanding nor compressed. Trend persistence elevated; momentum structure. Strong conviction from aligned signals.",
  "timestamp": 1712764800,
  "coin": "BTC",
  "timeframe": "1h",
  "also_see": [
    { "coin": "ETH", "timeframe": "1h", "confidence": 82 },
    { "coin": "SOL", "timeframe": "15m", "confidence": 73 }
  ],
  "_algovault": {
    "version": "1.23.0",
    "tool": "get_trade_call",
    "compatible_with": ["crypto-quant-risk-mcp", "crypto-quant-backtest-mcp"]
  }
}

The _algovault block makes outputs composable: downstream risk and backtest tools accept the object directly.

Live, verifiable track record

The full record is public and on-chain — no cherry-picking, no survivorship bias:

• Live dashboard — algovault.com/track-record
• Merkle batch verifier — algovault.com/verify
• Anchor contract on Base L2 — 0x6485…0f81
• All batches (raw JSON) — api.algovault.com/api/merkle-batches

AlgoVault is also a verified agent on the ERC-8004 Identity Registry (Base L2), agentId 44544 — a portable, on-chain handle AI orchestrators can resolve.

Works with your stack

AlgoVault is drop-in for every MCP-spec client, every major agent framework, and every official exchange Agent Trade Kit — no SDK, no wrapper. It serves Streamable HTTP at https://api.algovault.com/mcp; tools/list + resources/list is the API surface.

MCP clients.

Agent frameworks. First-party tutorials pair AlgoVault with each framework's canonical MCP adapter — copy-pasteable demo code, no SDK.

Each demo is runnable as python examples/<framework>/demo.py BTC 4h — gets a real BUY/SELL/HOLD verdict from api.algovault.com/mcp, prints it. ≤5 minutes to first call.

Exchange Agent Trade Kits. AlgoVault returns the analytics; your agent's risk policy decides what to execute. All demos run testnet/demo only.

Tools & resources

The MCP tools live at https://api.algovault.com/mcp. Every asset works across the full supported timeframe range, on major crypto perpetual venues.

• get_trade_call _{(alias get_trade_signal)} — composite BUY/SELL/HOLD verdict with confidence + regime, any asset, any timeframe.
• scan_trade_calls — scans the top-N perps by open interest on a venue; returns every actionable call in one shot.
• scan_funding_arb — cross-venue funding-rate spreads across 7 venues (Hyperliquid, Binance, Bybit, Gate, KuCoin, Aster, OKX), ranked. The only MCP server doing multi-exchange derivatives arbitrage.
• get_market_regime — classifies TRENDING_UP / TRENDING_DOWN / RANGING / VOLATILE for strategy selection.
• search_knowledge + chat_knowledge — free BM25 search and grounded Q&A over the full knowledge bundle.

Performance is exposed as a read-only MCP resource: performance://signal-performance (aggregated PFE win rate, never raw outcomes). Full parameter reference at algovault.com/docs.

Skills (20 ready-to-use Anthropic Agent Skills)

Single-prompt wrappers over 1–3 tool calls — regime gating, multi-timeframe consensus, funding-arb monitoring, and more. Install with claude plugin install AlgoVaultLabs/algovault-skills; browse at algovault.com/skills.

Pricing

Quota-only tiers. Every tier gets all venues, all assets, all timeframes — you pay for call volume, nothing else. HOLD verdicts are always free.

* HOLD verdicts (engine says "don't trade") are always free across all tiers — no x402 charge, no quota deduction. We only get paid when we see a tradeable opportunity.

Subscriptions: Sign up at api.algovault.com/signup. Starter ($9.99/mo) unlocks 3,000 calls/mo. API key delivered instantly after checkout.

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.

What's new in v1.23.0

• 📡 Cross-venue funding arbitrage now spans 7 venues. scan_funding_arb expanded from 3 to 7 — Hyperliquid, Binance, Bybit, Gate, KuCoin, Aster, and OKX — with interval-correct annualization per venue and a per-leg liquidity filter, so only tradeable spreads surface.

MCP clients cache tools/list at session start — toggle the connector off/on (or restart the MCP connection) to pick up the updated scan_funding_arb.

v1.22.0 highlights (recap)

• 🌐 <span data-tr-field="exchange_count">12</span> exchanges live. ASTER, BingX, Gate, HTX, KuCoin, MEXC, and Phemex join Hyperliquid, Binance, Bybit, OKX, and Bitget on the verified, Merkle-anchored public track record — composite trade calls, market-regime detection, and the per-venue PFE leaderboard now span every promoted venue.
• 🏆 12-venue leaderboard. The live track-record leaderboard now ranks per-venue PFE win rate across all 12 — filter to any newly-added venue.
• 🔁 Forward-stable coverage. The venue count reads live everywhere it appears, so coverage stays accurate as we keep adding venues.

Refresh your MCP client to pick up this release. MCP clients cache tools/list at session start — Claude.ai/Desktop: toggle the connector off+on; Cursor/Cline: restart the MCP server connection.

v1.21.0 highlights (recap)

• 📊 scan_trade_calls ranking lenses (rankBy) — rank the scan universe 9 ways (funding, gainers/losers, volume, open interest, volatility, 24h OI change) + aliases.
• 🧾 Verdict receipts (_receipts) — inline conviction, top factors, live track record, and a verify link on every verdict.
• 🔌 Gemini, Kraken & Alpaca pairings · 🏆 live track-record leaderboard · 🎁 referral program · 💸 pay-per-call on the trade-call & scanner routes.
• 🔒 Security hardening — x402 per-route price binding + replay protection; webhook egress-IP pinning + timestamped HMAC signatures. Breaking for webhook subscribers — update per docs/WEBHOOKS.md.

v1.20.0 highlights (recap)

• 📡 scan_trade_calls — market-wide scanner. One call scans the top-N perps (1–100, ranked by open interest) on your chosen venue and returns every actionable BUY/SELL with confidence and regime. HOLDs stay free — quota counts only actionable calls.
• 🔔 Webhooks: dynamic top:N watchlists. Subscribe with assets: ["top:25"] and your webhook follows the venue's top perps automatically. No manual list upkeep.
• 🏛️ TradFi-aware analysis. Both core tools now report the underlying market session (underlying_session, with weekend/holiday caveats), interpret fixed pre-IPO funding correctly, and aggregate cross-venue funding sentiment for stocks, indices, commodities, and FX across all 5 venues.
• 🧭 Smarter errors on young listings. Insufficient history returns a structured INSUFFICIENT_CANDLES error with suggested_timeframes instead of a plain string.

MCP clients cache tools/list at session start — toggle the connector off/on (or restart the MCP connection) to see scan_trade_calls.

Privacy

Local mode: zero telemetry — call history stays on your machine. Remote mode: request metadata logged for analytics (IP hashed, never stored raw). See the privacy policy.

License

MIT

Disclaimer: AlgoVault provides directional entry interpretation for AI agents. Exit timing is determined by your agent or strategy. This is not financial advice. Past performance does not guarantee future results.