Related Servers

Despite descriptions that frequently police each other ('prefer X over Y', 'use this instead of...'), the set is riddled with near-duplicates: get_actor_components vs get_actor_references, get_blueprint_components vs get_blueprint_components_list vs get_blueprint_scs_tree, configure_gravity vs set_world_gravity vs set_physics_gravity, get_undo_state vs can_undo vs can_redo, and four DataTable row readers with subtly different fidelity. An agent can usually stumble onto the right one because the prose is extremely explicit, but selection is a real cost, not a non-issue, across 2000 tools. Overlapping purposes are pervasive rather than occasional.

The dominant pattern is snake_case verb_noun (get_actor_label, set_actor_folder, add_light), but it is layered with many domain prefixes (interchange_, niagara_, pcg_, usd_, metasound_, anim_, rig_, hlod_, eqs_) and abbreviations (bp_, gc_, rvt_, bt_, ism_, mid_, mcp_). A few names are bare words with no verb (batch, region_start, bookmark), and reading-direction pairs like get_ vs has_ vs list_ vs read_ are used inconsistently for the same question. Readable, but a mixed convention rather than a single predictable one.

2097 tools is an extreme mismatch under any reading of the server's scope; even accounting for Unreal's breadth, dozens of near-duplicate readers/writers per subsystem inflate the surface far beyond what is needed. This is well past the '50+ tools' ceiling, and the count itself is the primary coherence problem.

Coverage is close to total: CRUD, lifecycle, read-back oracles and edge-case handling exist for actors, Blueprints, materials, Niagara, PCG, StateTree, Sequencer, USD and more, including deliberate honest gates where the engine has no headless path. A handful of genuinely acknowledged dead ends exist (set_preserve_scale_ratio does nothing, list_mass_signals always empty, several delegate/drag-drop gaps), but these are documented rather than surprising.