MCP stdio bridge for Blockout — drive the Blockout previs desktop app (stage scenes, choreograph cast & camera, export motion-reference packages) from Hermes, Claude Code, Codex, or any MCP client.
Blockout MCP demonstrates solid tool design with consistent naming (verb_noun pattern), comprehensive parameter descriptions, and well-structured JSON schemas. All 18 tools have explicit descriptions (avg 180 chars) and input schemas with typed properties. However, output schemas are entirely undocumented, LLMs cannot predict what fields to expect from responses, breaking the tool-chaining pattern. Error handling is absent; no guidance on recovery, retryability, or actionable error messages. Tool descriptions lack dependency hints (e.g., 'Call get_state first') and some parameters could benefit from tighter constraints (e.g., enum for format in export_motion_reference). Security posture is unclear, no evidence of permission gates or audit logging for destructive operations (delete_entity, clear_camera_marks).
Drop a timeline mark for an actor entity: at time t (seconds) the actor is at x,z. Chain marks to choreograph a walk. gait sets the movement style between marks.
Drop a camera mark at time t: camera at x,y,z (meters, y is height) looking with panDeg (clockwise from above, 0 faces -Z) and tiltDeg (positive looks up, negative down). focalLength in mm on Super 35 (24 wide, 35 normal, 50-85 tight).
Place a new entity on the ground. x/z in meters (+X right, -Z away); rotationDeg is clockwise from above with 0 facing -Z. Returns the new entity id.
Lay a full motion path (with altitude) on an entity from its current pose: plane-takeoff, heli-orbit, car-chase-weave, debris-fall… Replaces the entity's existing marks.
Apply a classic camera-move preset to the active shot camera: generates a full set of camera marks over the shot built around a subject, riding along if the subject moves (e.g. drone-follow a flying plane). Track-style moves also enable aim-lock.
Output schemas completely undocumented. No tool documents what fields it returns, breaking tool-chaining and forcing LLMs to guess downstream parameter types.
No error handling guidance. Destructive tools (delete_entity, clear_camera_marks) lack confirmation patterns, recovery hints, or retryability classification.
Missing dependency hints in descriptions. Tools like add_actor_mark, add_camera_mark, and apply_framing should document that get_state must be called first to obtain entity IDs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Auto-frame the camera using a preset relative to the labelled subjects: 2S (two-shot), OTS (over-the-shoulder), REV (reverse), TOP (top-down), LOW (low angle), DUTCH (canted).
Remove all camera marks from the active shot (keeps actor marks).
Remove an entity from the scene.
Export a motion-reference package (video + metadata) for the active shot: renders the scene from the camera path, bakes all entity motion into the video, and packages it with a JSON sidecar (entity positions, marks, timing). Saves to ~/blockout-export/[shot-name].zip.
Call FIRST. Returns a summary of the current project, scene, and active shot: the placed entities (id, asset, label, position) and the choreography marks on the timeline (actor + camera). Coordinates are in meters, +X is right, -Z is forward/away from the default camera; heading 0 faces -Z; rotationDeg is clockwise seen from above.
List motion-path presets for non-character performers: plane takeoff/landing/flyby, helicopter orbit, bird swoop, car chase moves, falling debris, thrown objects. Call before apply_action_preset.
List the placeable asset catalog (people, animals, vehicles, furniture, environment kits, primitives). Use an assetId from here with add_entity. Optionally filter by category.
List the classic camera-move presets (orbits, cranes, drone follows, vertigo dolly-zoom, whip pan…) with ids, categories, and descriptions. Call before apply_camera_move.
List the styles available per sequence type for spawn_sequence (dance styles, fight formats, chase modes).
Reposition an existing entity. Omitted fields are left unchanged.
Create a new shot in the current scene (same blocking, fresh camera). Returns the new shot.
Update the active shot settings. Omitted fields are unchanged.
Stage a whole choreographed crowd in one call: N dancers performing together, a paired brawl or mob fight, a foot chase, or a car chase. Creates the performers AND their choreography. Position with x/z/headingDeg (heading 0 faces -Z).
Parameter constraints could be tighter. export_motion_reference format is enum-constrained (good), but set_shot aspect ratio accepts free-form strings, should be enum or regex pattern.
No evidence of permission gates or audit logging for sensitive operations. Destructive tools (delete_entity) and export operations lack security context.