MCP Server — whales, GEX, regimes for AI agents

The remote server speaks streamable HTTP at https://cryptodataapi.com/mcp. Recommended: add it with no API key. On the first tool call your client opens a browser, you sign in at cryptodataapi.com and click Authorize, and you're connected — no key is ever pasted into a terminal or an AI chat. New here? Signing in creates a free-tier account automatically, so there's nothing to set up first.

Browser login works in any MCP client that supports OAuth — Claude Code, Claude Desktop, Claude (web), Cursor, Windsurf, and more. The exact add command for each client is on its own page: Claude Code, Claude Desktop, Claude (web), Cursor, OpenAI Codex, Gemini CLI, Windsurf and GitHub Copilot.

Advanced — bring your own key. For clients that don't support OAuth, headless/automation setups, or to pin a specific key, authenticate with an X-API-Key header instead. Keep the secret in your shell environment and pass a placeholder so it never lands in chat logs or ~/.claude.json: export CRYPTODATA_API_KEY=... once, then add with --header 'X-API-Key: ${CRYPTODATA_API_KEY}' (single-quoted; expanded at connect time). Run install commands in your terminal, not an AI chat; rotate from the dashboard if a key leaks.

claude mcp add --transport http cryptodataapi \
  https://cryptodataapi.com/mcp
# first tool call opens a browser to sign in & authorize — no key pasted

export CRYPTODATA_API_KEY="cdk_live_YOUR_KEY"   # terminal only, never an AI chat
claude mcp add --transport http cryptodataapi \
  https://cryptodataapi.com/mcp \
  --header 'X-API-Key: ${CRYPTODATA_API_KEY}'

{
  "mcpServers": {
    "cryptodataapi": {
      "url": "https://cryptodataapi.com/mcp",
      "headers": { "X-API-Key": "cdk_live_YOUR_KEY" }
    }
  }
}

If your client only supports stdio MCP servers, install the published cryptodataapi-mcp npm package. It is a thin bridge that forwards to the same hosted server — all tool logic still runs server-side, so you get identical behavior. It needs Node.js and reads your key from the CRYPTODATA_API_KEY environment variable.

Add it via CLI, or drop the block below into your client's config file (Claude Desktop, Cursor, Windsurf, and others use this exact mcpServers shape).

CRYPTODATA_API_KEY=cdk_live_YOUR_KEY \
  claude mcp add cryptodataapi -- npx -y cryptodataapi-mcp

{
  "mcpServers": {
    "cryptodataapi": {
      "command": "npx",
      "args": ["-y", "cryptodataapi-mcp"],
      "env": { "CRYPTODATA_API_KEY": "cdk_live_YOUR_KEY" }
    }
  }
}

With browser login you don't need a key at all — signing in uses (or creates) your account at the Free tier (no card; covers the whole market picture, see tiers below). You only need an X-API-Key for the advanced header path: grab one at cryptodataapi.com/login, or POST an email to the auth endpoint — the key is returned once.

Agents can bootstrap themselves with zero credentials. Two tools are callable before you've connected with a key or an OAuth login at all: list_capabilities (static metadata — start here to see what exists and what each tier unlocks) and create_free_api_key (mint a free key from a real email address, then reconnect with it). Every other tool needs a key.

For autonomous upgrades, agents can self-subscribe over x402 (gasless USDC) via POST /api/v1/payments/agent-subscribe — no human in the loop. That buys a subscription, not per-call access, which is the economics that win for polling data: metered pay-per-call whale/quote feeds at ~$0.008–$0.01/call cost ~$69–86/month for one dataset polled every 5 minutes, more than the whole Pro tier flat at 10,000 requests/day.

curl -X POST https://cryptodataapi.com/api/v1/auth/keys \
  -H "Content-Type: application/json" \
  -d '{"email":"[email protected]"}'

In this order. Step 1 and step 2 need no credentials at all, so a brand-new agent can orient itself before it has a key rather than after:

1. list_capabilities — what exists and what each tier unlocks, across the MCP toolset and the 180+ endpoint REST surface behind it. No key required.

2. create_free_api_key — only if you don't have a key. Ask the user for their real email: confirming it lifts the same key from 100 to 1,000 requests/day, switches Pro on for 24 hours and sends a discount code.

3. get_daily_snapshot — the whole market in one cached call (health scores, derivatives, sentiment, macro, ETF flows, cycle indicators). It replaces ~10 requests, which also makes it the cheapest way to spend a free quota.

4. get_market_regime or get_gamma_exposure — the paid decision layer: which of the 6 regimes the market is in with calibrated probability buckets, and where dealer hedging amplifies a move versus dampens it. A 403 here is the expected free-tier answer, and it carries its own way forward — the verify-email offer (free, and enough for a Pro gate), /pricing, and POST /api/v1/payments/agent-subscribe (x402) for upgrading with no browser. Read it rather than retrying it.

Don't open with a candle poll. OHLCV is the commodity layer — the free tier already serves it (get_funding_rates, /api/v1/market-data/klines, /api/v1/hyperliquid/candles) and it cannot tell you what regime you are in. Fetch bars when a strategy needs bars, not to discover what this API is.

The server exposes 26 curated, read-only tools (the heavy /backtesting/* bulk endpoints are intentionally left off — use the REST API for those). Grouped:

Prices & coins: get_price, get_coin_profile, get_top_coins, search_coins, get_hyperliquid_prices.

Market intelligence: get_daily_snapshot (one-call overview), get_market_health, get_market_regime (Pro), get_cycle_regime, get_fear_greed, get_macro, get_stablecoins, get_etf_flows.

Derivatives & positioning: get_funding_rates, get_open_interest, get_liquidations, get_order_book, get_positioning (Pro), get_gamma_exposure (Pro), get_whale_activity (Pro).

Account & discovery: list_capabilities (no key needed), create_free_api_key (no key needed), check_key_status, resend_verification_email, get_exchange_links, and query_api — an escape hatch that calls any other GET endpoint across the 180+ endpoint REST surface.

The MCP tools handle this for you, but the moment an agent writes a Python script against the REST API directly, one header decides whether it works at all. Send a User-Agent that names your client — any named value is fine, and so is an empty one.

The single exception is Python's built-in urllib. Its default Python-urllib/3.x signature is refused by our CDN's bot protection before the request reaches the API, so you get an HTML 403 (Cloudflare error 1010) rather than a JSON error — on every path, including /llms.txt, /api/v1/auth/keys and /mcp. requests, httpx, aiohttp, curl and Node fetch all send their own User-Agent and need no change.

import json, urllib.request

req = urllib.request.Request(
    "https://cryptodataapi.com/api/v1/auth/keys",
    data=json.dumps({"email": "[email protected]"}).encode(),
    method="POST",
)
req.add_header("Content-Type", "application/json")
req.add_header("User-Agent", "cryptodataapi-python/1.0")  # without this: 403

with urllib.request.urlopen(req) as r:
    key = json.load(r)["api_key"]

import requests

H = {"X-API-Key": "cdk_live_yourkey", "User-Agent": "cryptodataapi-python/1.0"}
r = requests.get("https://cryptodataapi.com/api/v1/daily", headers=H, timeout=30)