quant-mcp

Auditable ETF analytics
for your AI client.

A Model Context Protocol server that exposes UCITS ETF analytics — compare, factor metrics, Monte Carlo projection — as tools Claude, Cursor, Continue, or any MCP-compatible host can call. Every numeric output ships with its formula and source URL.

What you get

Four tools, one MCP server.

ping

Liveness check — confirms the server is reachable from your client.

compare_etfs

Two ISINs in, side-by-side metadata + OCF delta + AUM ratio + same-index check out.

get_etf_factors

1y / 3y / 5y / 10y annualised return, volatility, Sharpe, max drawdown, Calmar.

monte_carlo_projection

p10 / p50 / p90 percentile bands across N years for a weighted ETF basket.

Why this exists

Most AI clients can't talk to ETF data with formulas attached.

quant-mcp is a thin Model Context Protocol server that wraps a curated UCITS dataset plus the standard finance formulas — Sharpe, drawdown, Calmar, weighted Monte Carlo — and exposes them as tools any MCP-compatible client can call.

Every output ships with the formula that produced it, links to issuer factsheets, and a disclaimer. No subjective language. No "buy / sell / hold". Data only — you decide what to do with it.

Install

Install in 60 seconds.

Requires Python 3.12+ and uv. Clone the repo, sync deps, then point your MCP client at the directory.

# clone & sync dependencies
git clone https://github.com/mihaiandrei007/quant-mcp.git
cd quant-mcp
uv sync

Claude Desktop

Edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/ · Windows: %APPDATA%\Claude\) and add:

{
  "mcpServers": {
    "quant-mcp": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/quant-mcp", "run", "main.py"]
    }
  }
}

Claude Code · Cursor · Continue · Zed

Same config schema — drop the mcpServers block above into your client's MCP config file (e.g. .mcp.json for Claude Code). See full install docs.

Restart your client → run /mcp → confirm quant-mcp ✓ connected with 4 tools.

Try it

Paste this into your AI client.

use monte_carlo_projection with portfolio
[{"isin": "IE00BFMXXD54", "weight": 0.6},
 {"isin": "IE00BK5BQT80", "weight": 0.4}],
years 10, n_sims 2000

Projects a 60/40 VUAA / VWCE basket forward 10 years across 2 000 simulated paths. Returns p10 / p50 / p90 percentile bands per year + terminal multiplier + both disclaimers.

Compliance

Every payload, every tool.

Companion pack (paid)

Want the full CLAUDE.md + Cursor rules?

Building Python quant code with Claude Code or Cursor? The Claude Code Pack — Python Quant Edition contains the conventions this server was designed to integrate with: CLAUDE.md template, Cursor .mdc rules, finance formula reference, and a pre-wired MCP config pointing at this server.

€9 · 7 files · single-user license · lifetime v1.x updates. This OSS server stays fully free and MIT regardless — the pack just saves the rule-writing.

v0 is free

Twelve curated UCITS. MIT licensed.

No accounts, no auth, no billing. Install it, use it, file an issue if something breaks.