Binance MCP + ChainVector

One MCP server that pairs Binance execution with ChainVector market intelligence. Connect it to any MCP client (Claude Desktop, Cursor, Windsurf, ...) and the agent can read live signals, regime classification, probabilities and risk gauges from ChainVector, pick a strategy template, and execute the resulting decision on Binance — with intelligence and execution kept cleanly separated.

• Execution (Binance): the full Binance MCP tool surface — spot trading, TWAP/VP algo orders, convert, wallet, simple earn, staking, copy trading and more.
• Intelligence (ChainVector): 32 signal tools covering momentum, regime, risk index, probability engine, order flow, whale pressure, funding, liquidations, prediction markets and options context.
• Strategy templates: 7 selectable advisor tools that fuse ChainVector signals into structured trade decisions (long / short / flat, EV score, vetoes, size multiplier, stop/target).

Strategy advisors never place orders. They return a decision; placing the trade is always a separate, explicit Binance tool call.

Strategy templates

Ask your agent to run chainvectorStrategyList, or call a template directly:

Template	Tool	Setup it trades	Best regime
advise	chainvectorStrategyAdvise	Full ensemble: regime gate + EV weighting + veto stack	any (recommends the specialist template)
momentum	chainvectorStrategyMomentum	Momentum continuation with breadth + book alignment	trending
meanrev	chainvectorStrategyMeanRevert	RSI / Bollinger extreme back toward the mean	choppy / quiet
squeeze	chainvectorStrategySqueeze	Crowded funding side unwinding	squeeze_setup
cascade-fade	chainvectorStrategyCascadeFade	Fade an exhausting liquidation cascade	cascading (exhaustion)
carry	chainvectorStrategyCarry	Delta-neutral funding-carry venue pairs	quiet / any
probability-target	chainvectorStrategyProbabilityTarget	Quantified TP/SL plan: P(target before stop), expectancy in R	any (position planning)

Every decision includes the underlying signals, active vetoes (risk index extreme, cascade against direction, crowded squeeze against direction, cascading regime, no-momentum quiet regime) and warnings (venue dispersion, weak breadth, counter-trend). Treat action: "flat" or any veto as a hard no-trade.

Example prompts once connected:

• "Run the strategy list and tell me which template fits the current BTC regime."
• "Run chainvectorStrategyAdvise on BTCUSDT with a 60 minute horizon. If it says long with no vetoes, buy $100 of BTC at market."
• "Plan a long on ETHUSDT with take profit 2%, stop 1%: use the probability-target template and only proceed if expectancy is positive."

ChainVector signal tools

Core signals — chainvectorSignalsSnapshot (one-call decision snapshot), chainvectorMomentum (cross-venue scorecard, EV weight or veto), chainvectorLeadLag, chainvectorRegime (trending / choppy / quiet / squeeze_setup / cascading), chainvectorRiskIndex (0-100 stress circuit breaker), chainvectorSignalsHistory, chainvectorVolatility (vol battery + expected move), chainvectorIndicators (RSI/EMA/MACD/ATR/Bollinger), chainvectorProbability (six-estimator ensemble: terminal / touch / range, 30s-24h horizons, strike ladders).

Order flow & microstructure — chainvectorOrderflowCvd, chainvectorLongShort, chainvectorWhalePressure, chainvectorOrderbookImbalance, chainvectorOrderbookWalls, chainvectorPositioning.

Derivatives context — chainvectorFundingCurrent, chainvectorFundingWeighted, chainvectorFundingArbitrage, chainvectorFundingSqueeze, chainvectorBasis, chainvectorOpenInterest, chainvectorLiquidations, chainvectorCascadeRisk, chainvectorLiquidationHeatmap.

Prediction markets & options — chainvectorPredictionsMarkets / Quotes / Trades / Results / Stability (bid-price stability score) / Edge (model vs market repricing), chainvectorOptionsMaxPain, chainvectorOptionsImpliedProbability.

Requirements

• Node.js v18 or later (native fetch is used for ChainVector)
• A Binance API key + secret (create one — see readme/ for step-by-step screenshots)
• A ChainVector API key (chainvector.com — keys look like cv_live_...)

Installation

git clone https://github.com/C0inFlips/binance-mcp-chainvector.git
cd binance-mcp-chainvector
npm install
npm run build

Optional interactive setup (writes .env and offers Claude Desktop configuration):

npm run init

Configuration

Create a .env file (or copy .env.example):

BINANCE_API_KEY=YOUR_BINANCE_API_KEY
BINANCE_API_SECRET=YOUR_BINANCE_API_SECRET
CHAINVECTOR_API_KEY=cv_live_YOUR_CHAINVECTOR_KEY

MCP client config (Claude Desktop / Cursor / Windsurf)

Claude Desktop config lives at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
    "mcpServers": {
        "binance-mcp-chainvector": {
            "command": "node",
            "args": [
                "/absolute/path/to/binance-mcp-chainvector/build/index.js"
            ],
            "env": {
                "BINANCE_API_KEY": "YOUR_BINANCE_API_KEY",
                "BINANCE_API_SECRET": "YOUR_BINANCE_API_SECRET",
                "CHAINVECTOR_API_KEY": "cv_live_YOUR_CHAINVECTOR_KEY"
            },
            "disabled": false,
            "autoApprove": []
        }
    }
}

Restart the client after editing. The same block works in .cursor/mcp.json for Cursor.

How the strategy loop works

1. Choose — chainvectorStrategyList, or run chainvectorStrategyAdvise and follow its regime.recommended_strategy.
2. Decide — run the template tool for your symbol. It fetches the ChainVector signals it needs and returns a structured decision.
3. Gate — only act when action != "flat" and vetoes / blockers are empty. Scale position size by size_multiplier.
4. Execute — place the order with the Binance tools (BinanceNewOrder for spot market/limit orders, or the TWAP/VP algo tools for large orders), using suggested.stop_pct / suggested.take_profit_pct for exits.
5. Repeat — decisions are point-in-time reads, not standing orders. Re-run the template before every new entry.

Rate limits & reliability

• ChainVector tool calls are cached for a few seconds server-side in this process, so repeated reads within one agent turn reuse the same sample.
• On a 429 the client backs off automatically (honoring Retry-After) and reports the cooldown in the tool error.
• Strategy advisors make several ChainVector calls per invocation — don't poll them in a tight loop.

Disclaimer

Model outputs surfaced through these tools (probability, regime, heatmap estimates) are informational only and not investment advice. Trading cryptocurrencies involves substantial risk. Never commit real API keys to a repository; use environment variables or your MCP client's env block.

License

This project is open-source under the MIT License.