Scored across 49 tools
Most tools have sharply distinct purposes, and descriptions explicitly cross-reference each other (e.g. pm_data_orderbook says 'use pm_quote for that'). A few clusters are close enough to risk misselection: pm_data_events vs pm_data_event vs pm_data_canonical, pm_data_sources vs pm_data_sources_health, and the three whale tools (pm_data_whales / whale_wallets / whale_wallet).
There is a readable scheme (pm_data_* for cross-venue reads, get_* for account reads, open_/close_/place_/cancel_ for writes), but it mixes verb-first (get_candles, list_open_orders) with noun-first (pm_quote, spot_quote, futures_quote, crypto_price, tradfi_references, instrument_markets). The mixture is legible but not a single predictable pattern.
49 tools is heavy for a server, well past the 25-tool threshold, and the surface is padded by many niche data endpoints (calibration, volume history, whale wallets, sources health, canonical identity). While many are individually distinct, the sheer count strains an agent's ability to select well.
Coverage is broad and deep: spot/futures/PM quoting, opening, closing, SL/TP, portfolio, positions, trades, ledger, arena, plus a rich research data layer. Minor gaps exist, such as no explicit close_pm_position (PM positions appear to settle via the worker), but core lifecycles are covered.