The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the MCP Server listing page.
Quantustik is a risk-first market-intelligence layer for humans and AI agents — calibrated probabilistic forecasts, transparent track records, and machine-readable decision support through REST and MCP. It forecasts S&P 500 price paths with a quantum-mechanics model (the Schrödinger equation and Feynman path integrals), layers a machine-learning classifier and a market-conditions read on top, and turns the result into risk-first BUY / WAIT / AVOID / EXIT signals — each with entry price, stop-loss, take-profit ladder and position sizing, plus calibrated 90% confidence bands and a live, auditable track record that includes the losses.
What the analysis gives you, per ticker and for the whole index:
Who it is for: AI agents and their users doing market research — analysts, self-directed investors, and anyone who wants honest, uncertainty-explicit market context inside Claude, Cursor, ChatGPT or any other MCP client. Bullish calls are deliberately rare and earned: the model prefers saying WAIT to flattering you.
Educational research only — not investment advice. Nothing here is a recommendation to buy or sell any security. Calibration is measured, not guaranteed; past model performance does not predict future results.
A thin stdio proxy to Quantustik's hosted MCP server. It lets MCP clients
that only support local stdio servers (Claude Desktop, Cursor, Windsurf,
and others) reach the live, keyless, streamable-HTTP endpoint at
https://quantustik.com/mcp.
This package does not implement an MCP server itself, run any model
locally, or cache any data. It is a small wrapper around
mcp-remote pointed at our
hosted endpoint. Every tool call is forwarded live to quantustik.com and
answered from the same data and quota system as the hosted endpoint.
If your MCP client supports remote streamable-HTTP servers directly, you
don't need this package at all — just point it at https://quantustik.com/mcp.
This wrapper exists only for stdio-only clients.
The full tool catalog lives at quantustik.com/developers.
Add to your MCP client config (e.g. claude_desktop_config.json):
This starts a local stdio MCP server that proxies to
https://quantustik.com/mcp. Anonymous use is allowed under a per-IP hourly
quota; no setup required.
Get a free key at quantustik.com/developers, then set it as an environment variable:
| Variable | Purpose | Default |
|---|---|---|
QUANTUSTIK_API_KEY | Sent as Authorization: Bearer <key> to raise the anonymous per-IP hourly quota. | none (anonymous) |
QUANTUSTIK_MCP_URL | Override the remote endpoint (mainly for local development against a non-production instance). | https://quantustik.com/mcp |
Any additional CLI arguments you pass are forwarded to the underlying
mcp-remote proxy (e.g. --debug).
npx @quantustik/mcp-server runs a small Node script
(bin/quantustik-mcp.js) that spawns
mcp-remote against
https://quantustik.com/mcp using the http-only transport (our endpoint
is streamable-HTTP, not SSE). mcp-remote speaks stdio to your MCP client
on one side and streamable-HTTP to our server on the other — this package
just pins the URL, transport, and optional auth header so you don't have to
remember them.
Server implementation (Python, FastMCP): the hosted endpoint's source lives
in mcplafed/autoevolve
(app/mcp_server.py, server.json). This wrapper repo only contains the
thin npm client shown above.
All signals, forecasts, and market data returned by the Quantustik MCP server are for educational and research purposes only and do not constitute personalized financial advice. Past model performance does not guarantee future results. Always consult a qualified financial professional before making investment decisions.
MIT — see LICENSE.