Institutional crypto intelligence, in your AI agent
Connect CRYPTYX to Claude, Cursor, or any MCP-compatible agent. Get 470+ metrics across 8 factor classes, and every prior instance of a move with its historical outcomes, all through a single OAuth connection.
https://mcp.cryptyx.ai/api/mcpPaste this into any MCP client. OAuth consent handles the rest — no API keys, no config file surgery.
Choose your client
Three-step install for the most popular MCP clients. All routes converge on the same OAuth flow.
Claude.ai
- 1Customize → Connectors → Add → Add custom connector
- 2Paste the URL below, name it CRYPTYX
- 3Click Connect — browser flow handles auth
Works with Claude Pro, Team, Enterprise, and Max plans.
Cursor
- 1Cmd+Shift+P → "Open MCP Settings"
- 2Add new MCP server, paste the URL below
- 3Grant OAuth on next agent request
Requires Cursor 0.42 or later for OAuth MCP support.
Any client
- 1Any MCP-compatible client can use the standard HTTP transport
- 2Paste the URL below into your client's MCP configuration
- 3OAuth flow bootstraps automatically on first tool call
Works with LangChain, LangGraph, AgentKit, Anthropic SDK, OpenAI SDK, custom Python/TS agents.
What to ask first
Copy any of these into your agent after connecting. Each hits a different capability and returns institutional-grade evidence.
- Asset study
Where does SOL stand against its own history, short term and long term?
- Episode study
Report where BTC's latest move stands against its own history, and the historical outcomes of prior instances like it.
- Your rule
My rule: add to ETH only when it is statistically oversold against its 1-year history. Report where it stands now and the historical outcomes of prior instances.
- Market scan
Which assets have the most extreme 30-day volatility against their own history right now?
- Backtest
Test buying BTC when its 30-day return is below −1.5σ: report forward returns at 7, 30 and 90 days against the base rate.
- Multi-factor
On ETH, test the forward returns when funding and open interest were both extreme at once, and report the sample size.
Every number is computed by the engine, with its evidence
Every reading arrives as a z-score against the asset's own history, and every historical claim with its sample size, base rate and t-stat. Small samples are labelled, and a plain disclaimer travels with every response.
- 470+ metrics across 8 factor classes, each scored against the asset's own trailing year, with daily history back to 2021
- Episode studies with sample size, t-stat and a random-entry test
- Backtests at every horizon against the asset's base rate
- On-demand backtests with tier-gated compute budget
OAuth 2.1 + PKCE + Dynamic Client Registration
No API keys in config files. Per-user scoped tokens. Revocable from account settings. Subscription-tier gated with structured error responses so agents can surface upgrade paths cleanly. Full audit trail. Delegated identity via Stytch Consumer.
Developer reference, OpenAPI spec, tool catalog, and OAuth flow deep-dive: /docs/mcp