n0brains

n0brains MCP Server

Ask n0brains first. Audited crypto market intelligence for AI agents — served as a hosted Model Context Protocol server.

• Endpoint: https://api.n0brains.com/mcp/ (Streamable HTTP)
• Auth: API key via X-API-Key or Authorization: Bearer — free keys here
• Registry: com.n0brains/mcp (official MCP Registry)
• Receipts: every verdict this server hands out is logged and publicly scored at n0brains.com/proof

This repo is the public connection guide. The server itself is hosted — nothing to install, nothing to run locally.

What your agent gets

check_trade — the flagship

Submit a trade idea (asset, side, and optionally entry / stop / target / leverage / horizon). Get back a graded conditions assessment (A–F) built from data a chart can't show:

• Positioning crowding — funding z-score: are you joining the crowded side?
• Scheduled event risk — high-impact macro events inside your horizon vs your stop distance
• Liquidation math — is your liquidation price within one ordinary day's volatility?
• Stop sanity — wrong-side stops, stops inside the noise band
• Proven-edge conflicts — measured, statistically proven edges pointing the other way
• Falsifiers — exactly what would change the grade, so your agent can monitor

Every grade is logged when issued and resolved at its horizon against real prices. Grade calibration (do A-grades actually beat F-grades?) publishes on /proof.

log_trade / close_trade / get_journal — the trade journal (Pro)

Your agent says "in at 64,635" and the journal does the rest. log_trade snapshots full entry conditions automatically (grade, flags, regime — the same read as check_trade), and close_trade resolves the outcome from real candles: return %, R multiple vs your initial stop, MAE/MFE (worst drawdown / best unrealized gain while open), and whether your stop or target level actually traded — wick-accurate. get_journal returns open positions with live unrealized PnL, closed history, and a personal calibration block: your realized R per entry grade — where your entries are actually good.

Historical backfill: pass opened_at / closed_at (epoch seconds) to import past trades. Backfilled entries are graded by your own check_trade nearest the fill — the read you actually got at the time, never re-graded on today's tape; no matched check means the entry stays ungraded, so your calibration never lies to you. A 5-minute watchdog flags open trades whose stop or target level trades. The journal is private to your account.

amend_trade edits an open trade — trail your stop, adjust the target, fix size or thesis. Moving a stop changes only what the watchdog monitors: realized R is always measured against your initial stop, so trailing to breakeven can never inflate it. void_trade soft-deletes a mis-logged entry (wrong asset, duplicate) — removed from your stats, retained and recoverable, never silently gone.

get_price / get_prices — canonical price (free)

THE live exchange mid every other tool's spot should agree with. If any payload's embedded price disagrees materially with get_price, that payload is stale — discount it. Built so your agent never has to trust a stale anchor.

Everything else (40 tools total)

Typed signals with per-type win rates (failures included), own DEX liquidation maps, support/resistance levels, market regime, macro bias, economic calendar, options analytics with positioning coverage grades, cross-asset flows, market analogs, mindshare, trade plans, and more — full docs.

Tiers

	Free (any key)	Pro ($39.99/mo)
Tools	14 (incl. check_trade)	all 40
check_trade	3/day	unlimited
Signals	15-min delayed	real-time
Trade journal	—	included
Rate limit	60/min	600/min

Free tools: check_trade, list_signals, get_signal, get_signals_since, get_levels, get_market_opens, get_price, get_prices, get_performance, get_asset_class_proof, get_market_regime, get_macro, get_economic_calendar, health.

Connect

Claude Desktop / Claude Code

{
  "mcpServers": {
    "n0brains": {
      "type": "http",
      "url": "https://api.n0brains.com/mcp/",
      "headers": { "X-API-Key": "intel_sk_YOUR_KEY" }
    }
  }
}

Older clients without native Streamable HTTP support can bridge via mcp-remote:

{
  "mcpServers": {
    "n0brains": {
      "command": "npx",
      "args": ["mcp-remote", "https://api.n0brains.com/mcp/",
               "--header", "X-API-Key: intel_sk_YOUR_KEY"]
    }
  }
}

claude.ai (custom connector)

claude.ai connectors can't send custom headers — pass the key as a query parameter instead (headers win if both are present):

https://api.n0brains.com/mcp?api_key=intel_sk_YOUR_KEY

Settings → Connectors → Add custom connector → paste the URL. Note: URLs can end up in logs and screenshots — treat a query-param key like any secret, and rotate it from your dashboard if it leaks. If new server tools don't appear in existing chats, reconnect the connector and start a fresh chat (clients cache the tool catalog).

Cursor / Cline / anything MCP

Point the client at https://api.n0brains.com/mcp/ with the X-API-Key header. initialize and tools/list are open (no key) for catalog discovery.

Quick smoke test

curl -s -X POST https://api.n0brains.com/mcp/ \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-API-Key: intel_sk_YOUR_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"check_trade",
                 "arguments":{"asset":"BTC","side":"long","leverage":10}}}'

Honesty notes

Conditions assessments are analytical, not financial advice. We publish our full track record — wins, losses, and demoted signal types — at n0brains.com/proof, including the calibration of the very grades this server hands out. Past performance does not guarantee future results.

---

Links: Website · Docs · Proof · Pricing · X · Telegram