@cryptyx/mcp-server

CRYPTYX — the conviction engine for autonomous crypto trading agents.

Institutional-grade crypto intelligence delivered to AI agents via the Model Context Protocol. CRYPTYX converts fragmented crypto telemetry into factor scores, signals, multi-factor backtests, and regime classifications — so your agent can form conviction, not just fetch data.

Not a data proxy. A quant research platform. 21 tools across 376 metrics, 8 factor classes, ~200 tracked assets, and a daily-updating signal registry.

Execution is complementary. Use CRYPTYX alongside exchange toolkits like OKX and Kraken: they execute, CRYPTYX decides.

Install

npx @cryptyx/mcp-server

Claude Desktop / Claude Code

{
  "mcpServers": {
    "cryptyx": {
      "command": "npx",
      "args": ["@cryptyx/mcp-server"],
      "env": {
        "CRYPTYX_API_KEY": "your-api-key"
      }
    }
  }
}

Environment Variables

The 6-step conviction loop

CRYPTYX is designed for a specific agentic workflow. Most tools map to a step in this loop:

DISCOVER  →  DEFINE  →  VALIDATE  →  SCAN  →  STORE  →  EXECUTE

1. DISCOVER — get_featured_metrics surfaces the current top-performing metrics by information coefficient (IC). Start here.

2. DEFINE — analyze_metric or analyze_metrics_composite lets the agent build a multi-factor thesis (e.g. "trend momentum z > 1.5 AND funding stress z > 2.0").

3. VALIDATE — The same tools return forward returns at 8 horizons (1d to 365d). The agent sees whether the thesis has edge, not just vibes.

4. SCAN — scan_metric_universe runs the validated thesis across ~200 assets on the latest day. Which assets match the conditions right now?

5. STORE — fork_signal persists the thesis as a new inactive signal variant. The daily pipeline will track it forever.

6. EXECUTE — CRYPTYX doesn't execute. Hand off to OKX, Kraken, or whatever execution layer your agent uses.

Tool reference (21 tools)

Factor discovery — the IP moat

The core value of CRYPTYX. These tools let the agent do real quantitative research against 376 metrics across 8 factor classes.

Signal engine — parameterised conviction

A signal is a persistent, versioned, parameterised thesis. CRYPTYX ships with a catalog of active signals and lets agents backtest, fork, and tune them.

Market intelligence — state of the universe

Execution context

CRYPTYX Challenge

An open, public leaderboard where AI trading agents compete using real CRYPTYX signals. Used by the community, and a great source of benchmarking context.

Factor classes

Scale & data freshness

• 376 metrics defined across 8 factor classes

• ~200 digital assets tracked daily (target: 500+)

• 8 horizons: 1d, 7d, 14d, 30d, 60d, 90d, 180d, 365d

• Daily pipelines:

  • Metrics: 01:20 UTC

  • Signals: 02:27 UTC

  • Evaluation scorecards: 02:45 UTC

  • Agent optimisation: 03:00 UTC

• 15-minute refresh for spot prices and order book snapshots

• Weekly data source discovery agent scans 12+ providers for new signals

Example prompts

Build a thesis from scratch:

Use CRYPTYX to find the top metrics by IC, build a multi-factor thesis combining trend momentum with funding stress, backtest it on BTC, then scan the universe for assets matching both conditions today.

Explain a signal firing:

What signals fired today? Pick the highest-confidence one and explain why it fired on that specific asset.

Fork and tune:

Fork the TR_MOMO_CONT_14D signal with a stricter t_thr of 1.2, backtest both versions over the last 90 days, and tell me which one has better IC.

Regime-aware position sizing:

For my top 10 composite assets, what's the current regime? Size positions inversely to volatility regime — larger in trending, smaller in volatile.

Links

• Homepage: cryptyx.ai

• Source: github.com/cryptyx-ai/cryptyx-mcp-server

• OpenAPI spec: cryptyx.ai/openapi.yaml

• AI plugin manifest: cryptyx.ai/.well-known/ai-plugin.json

• AI reference: cryptyx.ai/llms-full.txt

• Changelog: CHANGELOG.md

License

MIT