CrashTestYourStrategy — MCP Server

Remote MCP server for portfolio & trading-strategy stress diagnostics.

Free tier — no key, no signup. Descriptive, never advisory.

https://mcp.crashtestyourstrategy.ai/mcp

Listed in the official MCP registry as

`io.github.fnobbe/crashtestyourstrategy`. This repository is the public server card —

the service itself is closed-source; everything an agent (or a reviewer) needs to

verify behaviour is linked below.

Quick start

Claude Code:

claude mcp add --transport http ctys https://mcp.crashtestyourstrategy.ai/mcp

Generic MCP client config (streamable HTTP):

{
  "mcpServers": {
    "ctys": {
      "type": "streamable-http",
      "url": "https://mcp.crashtestyourstrategy.ai/mcp"
    }
  }
}

claude.ai: *Settings → Connectors → Add custom connector* with the URL above.

The free tier is rate-limited per IP (30 req/min). No account, no key.

What it does

An open diagnostics layer that confronts a proposed portfolio or trading strategy

with the failure modes an optimistic backtest ignores — hedge breaks (the 2022

stock-bond case), sequence-of-returns risk, regime blind spots, deflated Sharpe.

Free tier (16 tools)

Full tier (token via contact):

`tier2_stress_test` (stress any ticker with auto-calibration + realism gate),

`build_portfolio`, `portfolio_frontier`.

Response contract

Every response is a versioned envelope (`ctys-agent-v1` family) carrying:

• `grounding_summary` — a risk-focused factual statement of the omitted tail
• `revision_required` — a gate-capable signal (never a directive)
• `methodological_limitations` — required disclosure of what the diagnostic does NOT establish
• `data_through` / `data_staleness_days` — self-reported data freshness (EOD, refreshed daily)

`validation://` resources expose the realism trust layer: 18 measured stylized facts

per asset vs historical reference bands — the falsifiability surface. Recompute

locally; trust by independent reproduction.

Scope & compliance

Model-based scenario simulation. Descriptive, never advisory: no suitability,

timing, or ranking claim is made or implied — the agent decides what "suitable"

means. Out-of-sample limits are disclosed, not hidden (e.g. the GFC hedge-behaviour

test failed and ships as a documented limitation).

Links

• Capability declaration: (JSON: /interop.json)
• Methodology:
• Weekly market-regime outlook (citable, auto-generated):
• Failure-mode ontology:
• llms.txt:
• Smithery listing:

Maintainer: @fnobbe · Operator contact: