Scored across 30 tools
Almost every tool targets a distinct resource/action pair (market snapshot, candles, book, positions, orders, history, desk brief, pool status), and descriptions explicitly cross-reference siblings to steer selection (e.g. get_market vs list_markets vs get_candles). The few adjacent pairs — size_trade vs place_bracket, set_stop/set_target vs sweep_brackets, cancel_order vs sweep_brackets — are clearly separated by the text, though the sheer density leaves some residual risk of misselection.
All 30 tools use a uniform snake_case verb_noun pattern (get_market, list_markets, set_stop, close_position, join_pool, submit_desk_call, update_config), with verbs drawn from a small consistent vocabulary (get/list/set/place/close/cancel/refresh/join/leave). No camelCase or stylistic mixing anywhere.
30 tools is heavy for a single server, well above the comfortable 3-15 band, though the surface legitimately spans market data, account state, order lifecycle, desk research, pool membership and configuration. Each tool does seem to earn its place, but the count is large enough that discovery cost is real.
Coverage is strong: market data (snapshot, board, candles, depth, funding regime), account state (balances, positions, orders, fills/funding/transfers, risk limits), full trade lifecycle (size, bracket, move stop/target, leverage, close, cancel, sweep), plus desks, pool and config. Minor gaps exist by design rather than oversight — no partial closes, no order modification (only cancel), and no fund movement — which agents can work around.