How I screen for ratio spread opportunities in 30 seconds with an MCP server

#ai #claude #mcp #showdev

The setup

A "ratio spread" in options trading is when you sell N options at one strike and buy M options at another, where N != M. The classic 1x2 put ratio spread (sell 1 ATM put, buy 2 OTM puts) is a favorite of vol traders because it lets you express a view that downside skew is overpriced and gives you positive convexity if the market really crashes.

The hard part is finding candidates. Skew mispricings are the kind of thing you used to need a Bloomberg terminal + a custom IV-surface model + an analyst to surface. With a free remote MCP server I built called Helium MCP, you can do it in 30 seconds inside Claude Desktop or Cursor.

What the MCP exposes

Helium MCP is a thin wrapper over a per-symbol ML options pricing model. For any contract, it returns:

predicted_price - Helium's model fair value
prob_itm - probability of expiring in the money
options_data_date - freshness of the chain snapshot

The model is trained per-symbol on each ticker's own historical options data, so it makes different (and sometimes wildly different) calls than a generic Black-Scholes fit.

Setup (one line)

Add to your mcp.json in Cursor or Claude Desktop:

{
  "mcpServers": {
    "helium": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://heliumtrades.com/mcp"]
    }
  }
}

That's it. Free, no signup.

Finding a ratio-spread candidate

Inside Cursor, I asked the agent: "Pull AAPL option prices from Helium for May 15 expiry across $180/$195/$200/$205 strikes."

Real output, just now:

get_option_price('AAPL', 200, '2026-05-15', 'call')   ->  $20.64,  prob_itm 0.52
get_option_price('AAPL', 205, '2026-05-15', 'call')   ->  $20.69,  prob_itm 0.52
get_option_price('AAPL', 195, '2026-05-15', 'put')    ->  $0.06,   prob_itm 0.01
get_option_price('AAPL', 180, '2026-05-15', 'put')    ->  $0.02,   prob_itm 0.01

Look at the puts. The model thinks AAPL has effectively 1% probability of finishing below $195 by mid-May. The market is paying actual money for those puts (and a lot more for the deeper OTM ones - that's the skew).

If you believe the model, the deep OTM put side is the overpriced side. That's a textbook setup for a 1x2 put ratio spread:

Sell 1 near-ATM put (collect the rich premium the market is offering for a bearish view)
Buy 2 deep OTM puts (cheap insurance + tail-side convexity)

If the market chops sideways or rallies (the model's base case), all the puts expire worthless and you keep the credit. If the market crashes hard, the 2 long puts catch up to and exceed the 1 short put.

The point isn't this specific trade

The point is that screening this kind of structural mispricing went from "needs an institutional setup" to "ask Claude in plain English" the day MCP made it possible to expose model APIs to LLMs.

The same workflow applies to:

Calendar spread arbitrage (compare term-structure of Helium IV vs market IV)
Diagonal spreads (mix the two)
Volatility compression candidates - Helium MCP has a get_top_trading_strategies endpoint that returns a daily-ranked long-vol vs short-vol screen with explicit bull/bear cases:

"Market IV (~30%) above Helium IV (~26-27%) across maturities - favoring volatility compression. Skew is mostly tail-priced..."

The bigger picture

ML option pricing was an institutional moat for two decades. With MCP, the marginal cost of querying an ML option pricing model from inside an LLM is zero. That changes who can run a structured options screen.

Helium MCP also exposes 31-dimension structured bias scoring across 3.2M+ news articles (5,000+ sources) - useful for the news/narrative side of trading - but that's a topic for another post.

Caveats

This is not trading advice. Helium's model can be wrong. Per-symbol regression models can overfit. Always size positions appropriately and validate against your own framework.