v0.4.0 · Part of cexagent · MIT open source

The only Hyperliquid MCP
with a judgment layer.

22 tools · 60 top leaderboard traders pre-seeded · regime classifier + anomaly scanner + historical baselines + EN/ZH narrative · zero backend · MIT open source. Every other HL MCP wraps the REST API; we turn data into a read.

Install in 30 seconds →
stdio MCP · Node 20+ · zero user data · zero config
22
tools
live & tested end-to-end
60
traders pre-seeded
from HL's 34,684-row leaderboard
0
backend services
your agent talks to HL directly
0
user data stored
no accounts, no keys, no telemetry
Why it exists

A terminal built for agents, not humans.

01
We read HL's leaderboard so you don't have to
Hyperliquid publishes 34,684 trader addresses ranked by PnL. We pull the top 30 all-time + top 20 month + top 20 week — deduped, labelled with public displayNames (BobbyBigSize, jefe, Auros, 憨巴小龙, thank you jefef…). Any tool that needs 'addresses' just uses them. No Arkham subscription, no Etherscan digging.
02
Not just raw data — a read
22 tools across 4 modules. A regime classifier that labels every asset as 'organic rally' / 'short squeeze' / 'funding bleed' / 'coiling' from price + OI + funding. An anomaly scanner that distills 200+ perps into 3 headlines. Historical context that tells you whether +12%/yr funding is extreme or boring. Narrative briefings in EN & ZH.
03
Zero backend. We literally cannot see you.
The MCP server is a subprocess of your agent. Requests go straight to api.hyperliquid.xyz — nothing routes through our servers. We run no database, no analytics, no telemetry. No API keys for you to leak. MIT source on GitHub — diff it before you install.
Tool matrix

Four modules. One install. All open source.

Each module covers a different shape of HL intelligence. Ship progresses tool by tool — check the dots.

A6/7 ready
Liquidation Risk
  • get_top_liquidation_risks
  • liquidation_heatmap
  • simulate_cascade
  • my_position_risk
  • simulate_my_liq_price
  • get_recent_liquidations
  • historical_cascade_replay
B5/6 ready
Whales & Flow
  • whale_pnl_leaderboard
  • get_whale_flows
  • address_position_history
  • smart_money_flow
  • get_funding_pnl
  • new_whale_entries
C8/9 ready
Market Structure
  • get_funding_divergence
  • asset_snapshot
  • get_all_asset_ctxs
  • hlp_metrics
  • orderbook_imbalance
  • compare_perps
  • historical_context
  • detect_anomalies
  • insurance_fund_status
D3/3 ready
Narrative
  • explain_market_structure
  • asset_snapshot_narrative
  • daily_briefing
Example prompts

Just talk. No addresses, no config.

Q
“Are HL whales long or short BTC right now?”
→ smart_money_flow()
Q
“What's anomalous on HL right now?”
→ detect_anomalies()
Q
“Top 20 traders on HL this week by PnL.”
→ whale_pnl_leaderboard()
Q
“If BTC drops 5% how much gets liquidated?”
→ simulate_cascade()
Q
“Is BTC funding extreme vs its 30-day baseline?”
→ historical_context()
Q
“Give me a trader's read on HL market structure.”
→ explain_market_structure()
Honest comparison

We checked what else is out there.

There are ~5 other HL-themed MCP servers on GitHub. Here's exactly how we stack up — star count sourced today, every column independently verifiable.

projecttoolsanalysis layeraddresses pre-seededlast updated
hyperliquid-radar22regime classifier · anomaly scanner · historical baselines · EN/ZH narrative60 (live leaderboard)today
mektigboy/server-hyperliquid~5read-only wrappersnone2024
kukapay/hyperliquid-info~7read-only wrappersnone2025-mid
Br0ski777 HL pack~6 (fragmented)per-endpoint toolsnonemixed
Install

Paste one line. Restart. Done.

The stdio MCP server runs locally in your agent's process. Your queries hit Hyperliquid's public API directly — nothing flows through our servers.

~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "hyperliquid-radar": {
      "command": "npx",
      "args": ["hyperliquid-radar"]
    }
  }
}
Privacy

Nothing to collect. Nothing to leak.

privacy_policy.txt

The MCP server runs as a subprocess of your agent. Every tool call goes directly from your machine to Hyperliquid's public API. We run no backend. We see no requests. We have no user data to sell, leak, or subpoena. Source-available on GitHub, MIT license.