Model Context Protocol server for Hyperliquid — read-only public data (positions, fills, funding, PnL) for any wallet address.
Strong tool naming (all verb_noun pattern), comprehensive descriptions (avg 180 chars, well within 10-1024 range), and complete Zod schemas with type definitions. All 8 tools have descriptions and input schemas. Parameters are well-documented with enums where appropriate (e.g., get_top_traders has 'window', 'sort_by' as enums). However, output schemas are not explicitly documented in the source, only input schemas are visible. Error handling is basic (validateAddress checks, try-catch in hlInfo with retry logic) but lacks actionable recovery guidance for LLMs. No tool annotations (readOnlyHint, destructiveHint) despite all being read-only. Composition is clean: each tool has one responsibility, and tools chain well (e.g., get_top_traders returns addresses usable by get_positions).
Compare 2–5 Hyperliquid wallets side-by-side: account value, 30d PnL, win rate, avg ROI, top position.
Get recent fills (trades) for a Hyperliquid wallet. Returns up to 500 fills from the last N days (default 7).
Get funding payments/receipts for a Hyperliquid wallet over the last N days (default 7). Returns net funding, paid, received, and per-coin breakdown.
Compute realized PnL summary for a Hyperliquid wallet: total PnL, closed trades, win rate, avg ROI, best/worst trade, avg hold time.
Get open perpetual positions for a Hyperliquid wallet (size, entry, mark, uPnL, ROE, leverage, liquidation price) plus account value, margin used, withdrawable.
Get current market price for a Hyperliquid perp (mid, mark, oracle, 24h change, funding rate, OI, 24h volume).
Output schemas not documented. Tool descriptions state what is returned (e.g., 'size, entry, mark, uPnL, ROE, leverage, liquidation price') but formal output schema definitions are absent from source. LLMs cannot reliably plan downstream calls or extract fields without explicit schema.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All 8 tools are read-only and idempotent, but this is not declared in the tool registration. LLMs cannot infer safety properties and may incorrectly reason about retry semantics.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Get spot token balances for a Hyperliquid wallet (free, locked, total; USD value where available).
Return the top N Hyperliquid traders by PnL / ROI / volume over a time window (24h, 7d, 30d, or all-time). Uses Hyperliquid's own public leaderboard endpoint — no API key, covers all 30k+ accounts. Returns addresses you can feed into get_positions / get_pnl_summary / get_fills for deeper analysis.
Error messages lack recovery guidance. hlInfo() throws generic errors like 'HL API 500 Server Error' or 'Invalid Ethereum-style address'. LLMs receive no actionable next steps (e.g., 'retry in 5s', 'call search_wallets() to find valid address'). Errors should guide the agent toward resolution.
compare_wallets accepts 2 - 5 addresses but no validation enforces this constraint in the schema. The description states '2 - 5' but the input schema shows only 'array' with no minItems/maxItems. LLMs may pass 1 or 10 addresses, causing silent failures.
get_top_traders has complex filtering logic (exclude_zero_volume, exclude_market_makers, max_turnover_ratio) with interdependencies not fully documented. Parameter descriptions are thorough but lack explicit guidance on which combinations are valid or recommended.