Tokenized stocks trade 24/7 while their underlyings sleep. We built a real-time delta tracker and exposed it to AI agents over MCP — with a 402 paywall on Arc.
A delta tracker computes the price divergence between a tokenized stock and its underlying in real time and serves it to AI agents over MCP — free tier 1 req/min, then HTTP 402 with an x402 payment on Arc Testnet.
Picture this: Apple trades at $336.45 on Nasdaq. Tokenized Apple (AAPLB on Binance) trades at $336.44 at the same moment. A penny apart — noise. But on a Saturday night, with Nasdaq asleep, AAPLB can drift to −2%. That is not noise anymore. That is a signal.
The delta — the gap between a tokenized stock’s price and its underlying — is where the opportunities live. Humans cannot watch 20+ tokens around the clock. A machine can. So we built a tracker that computes the delta in real time and handed it to AI agents.
Binance streams token prices around the clock; Finnhub/Alpaca supply the equity prices. DeltaEngine computes the delta, the market phase (O/C/P) and the stale flag (15s without updates). The MCP server exposes it to the agent through five tools, and Telegram receives an alert the moment a delta crosses the 0.5% threshold.
What the tracker computes #
For every bStock the tracker keeps two prices:
- bStockPrice — the token’s price on Binance, where tokenized stocks trade 24/7;
- underlyingPrice — the real stock’s price from market data providers (Finnhub, with an Alpaca fallback).
Delta in percent:
(tokenPrice − stockPrice) / stockPrice × 100. Alongside it,
three flags a trader actually needs:
- phase — is the stock market open (
O), closed (C), or in pre-market (P); - stale — the price feed went quiet for more than 15 seconds, so treat the number with care;
- inAlert — the delta crossed the0.5% threshold (configurable).
A live response looks like this:
{"symbol":"AAPLB","underlying":"AAPL","multiplier":1.0006,
"bStockPrice":336.44,"underlyingPrice":336.45,
"deltaPct":-0.064,"phase":"O","stale":false,"inAlert":false}
How an agent uses it #
An AI agent is a program — Claude, a GPT-based bot, your own script — that calls services on its own. Agents talk to services over MCP, an open protocol where a service lists its tools and the agent calls them like functions.
The tracker is an MCP server with these tools:
get_delta— delta for one symbol;list_deltas— all symbols at once;get_quote— current quote;get_events— event history (threshold crossings, stale feeds);get_digest— the day’s summary.
curl -X POST https://agentbadge.xyz/mcp/bstock/tools/get_delta
-H "Authorization: Bearer $TOKEN"
-d '{"symbol": "AAPLB"}'
There is also Telegram: subscribe_telegram registers the
agent’s operator, and the tracker pushes a message whenever a delta
crosses 0.5%. No polling needed — the signal finds you.
Freemium: a snapshot for free, the stream for a fee #
The first request each minute is free. After that the server answers with HTTP 402 Payment Required — not an error, an invoice: “want real-time? pay”. The x402 standard lets the agent pay automatically: 5 USDC on the Arc network buys 30 days of access. No signup, no card, one on-chain transaction — gas paid in USDC too, an Arc specialty. The server verifies the transaction on-chain and opens the door. The same payment cannot be replayed — replay protection is built in.
Why a trader should care #
A −0.06% delta is noise. A −2% delta on a closed exchange means “the token was sold off and the equity has not woken up yet”. Whoever sees it first captures the convergence. A person cannot monitor every token every second — an agent can, and it pays for its own data feed without a human in the loop.
Next: inside the 402 paywall — how freemium works when the customer is a machine.
Links
- Agent guide (endpoints, limits, examples): agentbadge.xyz/bstock-guide
- All articles in the series: agentbadge.xyz/blog
- MCP endpoint:
https://agentbadge.xyz/mcp/bstock/tools/get_delta