🧠 Quant Brain MCP

Turn Claude into a quantitative analyst for US and Indian equities.

Ask in plain English. Get sized trade plans, portfolio optimization, backtests, and price alerts — grounded in real market data, not vibes.

You:    "Scan my watchlist and build a trade plan for whatever looks most actionable.
         ₹2,00,000 equity, 1% risk."

Claude: RELIANCE.NS flagged (at 20-DMA, volume 1.8x average).

        TRADE PLAN — RELIANCE.NS (long)
        Entry           ₹1,310.00
        Stop            ₹1,270.10   (swing low, 2.1 ATR)
        Size            50 shares   (₹65,500 — 32.8% of equity)
        Max loss        ₹1,995      (1.0% of equity)
        Targets         1R ₹1,349.90 · 2R ₹1,389.80 · 3R ₹1,429.70
        Invalidation    Thesis invalid below ₹1,270.10 — exit without debate.

No API keys. No accounts. Connect one URL and start asking.

⚡ Quickstart

Claude Desktop / Claude Web → Settings → Connectors → Add custom connector → Streamable HTTP:

https://mcp-quant-brain.onrender.com/mcp

That's the whole setup. Try: "What's RELIANCE trading at, and is it overbought?"

Free-tier note: the server sleeps when idle and takes ~50 s to wake. If the first request times out, retry once. Details in Getting Started.

Related MCP server: vibetrader-mcp

🛠 What you get — 25 tools

Full reference with every parameter: docs/tools.md

🔄 The workflow it's built around

flowchart LR
    S["🔍 <b>scan_watchlist</b><br/>what moved this week?"] --> N["📰 <b>get_news</b><br/>why did it move?"]
    N --> P["📐 <b>build_trade_plan</b><br/>entry · stop · size · targets"]
    P --> A["🔔 <b>price_alert</b><br/>watch my levels"]
    A -.->|level breaks → notified| P
    subgraph deeper["go deeper anytime"]
        I["📊 indicators"] ~~~ B["🧪 backtests"] ~~~ O["💼 optimizer"]
    end
    S -.-> deeper

🔔 Price alerts that survive restarts

You:  "Alert me if RELIANCE drops below ₹1,270"        → stored server-side (Postgres)
      ...
Bot:  "PRICE ALERT FIRED — RELIANCE.NS moved below 1270.00, now at 1268.20"

Alerts are one-shot, persist across server restarts, and pair with a scheduled Claude task that checks hourly during market hours and pushes to your phone. Setup in docs/price-alerts.md.

🎯 Why this instead of a stock screener?

1. It answers the trading question, not just the data question. Indicators tell you RSI is 43. build_trade_plan tells you entry, stop, how many shares, and where your thesis dies — sized to your account.

2. India is a first-class citizen. NSE tickers, NIFTY benchmarking, 8 Indian sector indices, ₹ formatting. Not a US tool with .NS bolted on.

3. The numbers are audited. Every calculation was adversarially tested against textbook references and live data — 161 automated tests pin the math, including regression tests for 11 real bugs found and fixed along the way. See docs/architecture.md.

4. Honest about its data. Delayed quotes are labeled with timestamps. Stale feeds are flagged, not hidden. FX limitations are disclosed, not papered over.

📚 Documentation

⚠️ Honest limits

• Data: Yahoo Finance. US quotes near-real-time; NSE/BSE ~15 min delayed. Daily bars for analysis.

• No options chains, futures, intraday candles, or tick data.

• Not investment advice. Educational analysis tooling. Every trade plan says so and means it.

License

MIT — use it, fork it, ship it.