No user-submitted related servers found.
Scored across 274 tools
While many names are individually descriptive, large blocks of tools share verbatim boilerplate descriptions (every M6 animation_* unit, every facial_* unit, every physics_* unit, every material_*/texture_* unit), so an agent must infer purpose from the name suffix alone. There is also genuine conceptual overlap: mesh_region_define / character_define_regions / character_lock_regions / mesh_region_verify / character_validate_locked_regions all manipulate and check region manifests, and mesh_candidate_create / character_create_candidate / retopology_create_candidate / physics_create_candidate all create candidates with nearly identical semantics described separately.
Names follow a stable snake_case pattern with a domain prefix (mesh_, character_, rig_, animation_, retopology_, uv_, bake_, material_, facial_, physics_) plus a verb_noun suffix, which is predictable and greppable. A few entries drop the verb (scene_audit, physics_plan, physics_status) and prefix families overlap slightly (uv_analyze vs uv_analyze_texel_density), but overall the convention holds.
274 tools is a massive mismatch for any single MCP server surface, far beyond what an agent can reliably select from. Many are near-duplicate lifecycle stages (analyze/propose/preview/apply/verify/revert) replicated per subdomain, inflating the count well past usefulness.
Coverage of the Blender production pipeline is genuinely deep: checkpointing, asset roles, region locking, retopology, UV/UDIM, baking, rig analysis, animation/mocap/facial, physics, materials and textures, each with QA gates and a plan-first mutation model. Gaps are minor (little direct scene-object/export manipulation), but the surface is arguably over-complete rather than incomplete.