# Kwant — Quant Intelligence for AI Agents

> Kwant is quant intelligence delivered as an endpoint. A hosted Model Context
> Protocol (MCP) server that lets an LLM agent reason about markets like a quant
> desk — data, signals, factor scores, screens, backtests, and portfolios over
> US & TSX equities. Agents query parameterized tools over Streamable HTTP and
> receive structured JSON. No install, no Python. Computation runs server-side on
> Cloudflare Workers.

Endpoint (MCP over Streamable HTTP, JSON-RPC 2.0): https://kwant.sh/mcp
Landing page: https://kwant.sh/
Health: https://kwant.sh/health

## How to connect

Every /mcp request needs a valid API key (`Authorization: Bearer kwant_…`) —
including the handshake. The auth gate runs before billing, so a key opens the
session; only `tools/call` is additionally charged.

1. POST `initialize` to /mcp — requires a key, no charge. Returns
   serverInfo {name: "kwant"} + an `mcp-session-id` header.
2. POST `tools/list` — requires a key, no charge. Returns the 16 tools with JSON
   schemas.
3. POST `tools/call` — requires a key AND payment. Charge is drawn from a prepaid
   balance funded by Stripe top-up, OR paid per query with an x402 `X-Payment`
   header (USDC on Base mainnet). An authenticated-but-unpaid call returns HTTP
   402 advertising every enabled rail (`Accept-Payment: x402, prepaid`).

Capability negotiation (`initialize`, `tools/list`, `ping`) carries no charge but
still requires a key. No key → HTTP 401. Keys are provisioned by Kwant.

## The three layers

Kwant turns raw market data into judgment across three ascending layers:

Data layer — $0.005 per query:
- get_quote: latest point-in-time quote for a ticker.
- get_price_history: OHLCV bars over an interval/range.
- get_fundamentals: profile + key ratios (market cap, PE, sector, …).

Signal layer — $0.005 per query ($0.010 for the correlation matrix):
- compute_indicator: RSI, MACD, SMA, EMA, BBANDS, ATR, ADX, STOCH.
- detect_signals: golden/death cross, RSI bands, MACD cross, breakout.
- compute_stats: volatility, sharpe, max_drawdown, returns, beta, correlation.
- compare_tickers: rank N tickers by a chosen metric.
- compute_correlation_matrix: pairwise correlation across a basket ($0.010).

Intelligence layer — $0.010 per query:
- screen: filter a universe by fundamental/technical rules.
- screen_with_scores: screen + rank with composite factor scores.
- build_monthly_universe: construct a point-in-time monthly universe.
- compute_universe_scores: factor scores across a whole universe.
- construct_portfolio: build a weighted portfolio from a selection.
- compute_portfolio_stats: portfolio-level risk/return statistics.
- backtest: backtest a signal/strategy on one ticker.
- run_portfolio_backtest: monthly-rebalanced multi-asset backtest.

## Tickers

US symbols are bare (AAPL). TSX uses the Yahoo `.TO` form (RY.TO) or the `TSX:RY`
prefix. TSX Venture uses `.V`. Indices use the caret form (^GSPC).

## Payment rails

- x402 (crypto): per-query USDC settled on-chain (Base mainnet) via the CDP
  facilitator. Send `X-Payment`; receive tx hash in `X-Payment-Response`.
- prepaid (fiat): fund a balance via Stripe Checkout (`POST /topup`), each query
  debits it. Read balance at `GET /balance`.

## Notes

- Data is for research and analysis; not investment advice.
- Prices above are authoritative and enforced server-side per tools/call.