dinoctor-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DINOCTOR_APP | No | Full path to the Dinoctor app, if it isn't in the usual place (e.g. some other location or the Microsoft Store version). | |
| DINOCTOR_HOME | No | Dinoctor's data folder, if you moved it. | |
| DINOCTOR_PORT | No | The port of a Dinoctor that is already open, if you changed it. Normally found by itself. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Dinoctor is a video app where you direct animated videos: a project has scenes, a scene has shots, and a shot has a place, a cast of characters and objects, actions, text, a camera move, lighting and a transition. Start with open_dinoctor, then get_project. Use library to see what each style offers before choosing ids. After changes, use look to see the shot and fix what's wrong. Tools that spend the person's credits return a price first and need confirm: true; always ask the person before confirming.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| open_dinoctorA | Start the Dinoctor app on this computer if it isn't open, and wait (up to about 90 seconds) until its window is ready. Call this first. Returns { opened, ready }. Fails with a download hint if the app isn't installed. To only check whether it's open, use status. |
| statusA | Quick read-only check: is Dinoctor open, the project's style and format, how many scenes and shots, and which shot is on screen (id and indexes). Returns { open: false, note } when the app is closed. For every scene and shot, use get_project. |
| get_projectA | Read-only: the whole project in short form. Style, format, and every scene and shot with id, index, length, camera, place, cast (member ids and refs), number of actions, texts and transition. Read it before editing to get the ids other tools need. For every setting of one shot use get_shot; for a picture use look. |
| get_shotA | Read-only: every setting of one shot as JSON (place, cast, actions, emotions, texts, camera and camera keys, lighting, transition, keyframes), with its scene and shot index. Use it to see exact fields and ids before edit_project or keys. For a short view of all shots use get_project; to see the shot as a picture use look. |
| go_toA | Show a shot in Dinoctor's editor so the person sees what you're working on. Changes only what's on screen, not the project. Edit tools use the shot on screen when no shot is given, so this also sets their default. Returns { showing }. To get a picture of the shot, use look. |
| edit_projectA | Low-level edit of any project setting by JSON Pointer path (read how_to once for the format, get_shot for a real shot). Each change: {"op":"replace"|"add"|"remove"|"merge","path":"/scenes/0/shots/1/seconds","value":4}. All changes in one call are one undo step; a bad path or value is refused and nothing changes. Returns { done: number of changes }. Prefer the specific tools (cast, act, words, camera, light, place, transition, keys); use this only for fields they don't cover. |
| new_projectA | Replace the open project with a new, empty one (one scene, one shot with the style's start place, light and lead character). The first call only says what will happen; ask the person, then call again with confirm: true. The person can Undo it right after. To keep the project and only change its look, use set_style; to open a saved file, use open_project. Returns the new scene and shot ids. |
| set_styleA | Switch the whole project to another built-in art style. Every shot is rebuilt from the new style's own place, light and lead character: actions, cast, scenery and keyframes go; texts and lengths stay. If the project already has work, the first call returns needsConfirm; ask, then call with confirm: true. One undo step for the person. For one of the person's own styles use use_style; to start fresh use new_project. |
| set_formatA | Make the whole video horizontal (16:9, YouTube) or vertical (9:16, Reels / Shorts / TikTok). Shots stay as they are, framed for the new shape. One undo step for the person. Returns { done } with the ids of what changed. |
| add_sceneA | Add a scene (a group of shots, usually a new place or moment) with one shot in it, and show it. The shot starts like the app's + Scene: the last shot's place, light and cast (or the style's plain place), a still medium camera, 3 seconds. Returns the new scene id, shot id and cast ids. One undo step for the person. To add a shot to an existing scene, use add_shot. |
| add_shotA | Add one empty shot (no cast, no scenery items; place, light and look carry on from the shot before, like the app's + Shot). By default it goes right after the shot on screen. Returns the new shot id. Then fill it with cast, act, words, camera, light, place and transition. One undo step for the person. For a new place or moment use add_scene; to copy a shot use duplicate. |
| removeA | Remove a shot or a whole scene by id. A scene's only shot can't be removed alone (remove the scene), and the last scene can't go. One undo step for the person. Returns { done }. To take a person or object out of a shot, use cast with op remove. |
| moveA | Move a shot to another position, in its own scene or into another scene. The shot keeps all its settings. One undo step for the person. Returns { done, shot }. To copy instead of move, use duplicate. |
| duplicateA | Copy a shot (the copy goes right after it) or a whole scene (after it). Copies are separate: changing one never changes the other. Returns the new ids. One undo step for the person. For a new empty shot use add_shot. |
| cut_shotA | Cut a shot in two at a time, like a clip in an editor: both parts together play exactly what the shot played; the second keeps the transition and the voice stays in the first. Times round to 0.1 s and must be at least 0.2 s from either edge. Returns { first, second } shot ids. One undo step for the person. To change only the length, use edit_project on seconds. |
| rename_sceneA | Rename a scene (the name the person sees in the scene list and the timeline). Names don't have to be unique and change nothing else in the film. One undo step for the person. Fails if the scene isn't found or the name is empty. Returns { done }. Shots have no names; scenes are listed by get_project. |
| castA | Add, change or remove a character or object in a shot. op add needs ref (a character or object id of the style); change and remove need member (its id from get_shot, or its name). Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). A character added without x is spaced out automatically. Returns the shot and member ids. One undo step for the person. To make the cast move or feel something, use act; for animation over time, use keys. |
| actA | Direct the cast of a shot. kind action: a move (type from library actions) by actor, some need a target; timed by when or exact at %. kind emotion: a face feeling on member at %. kind talk: member's mouth moves from % to %. kind look: member looks at target from % to %. op add (default), change (id), remove (id), clear (all of that kind). Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the item id. One undo step for the person. To place or remove people use cast; for exact paths over time use keys. |
| wordsA | Add, change or remove on-screen text in a shot. preset is one of the style's text presets (its look and font); action is its animation (pop, slam, type, fade, slide, drop...). Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the shot and text ids. One undo step for the person. For a script line that is spoken, use write_script or assign_line (its captions come from the voice). |
| cameraA | Set a shot's camera: a preset move with strength and wiggle, or direct it with camera keys (keys win over the move). Each key: at (seconds) and any of x, y, z, distance, height, angle, roll, ease; what a key leaves out is taken from where the camera is at that moment. Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the camera settings. One undo step for the person. For the end-of-shot camera move into the next shot, use transition (outro_camera). |
| lightA | Light a shot: a lighting preset (the style's moods), and/or the ambient colour and level, and single lights. Any hand change makes it a custom look. Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the lighting. One undo step for the person. To change the light over time, use keys (scene:light or light:). |
| placeA | Set where a shot happens: a place preset (with its scenery), and/or its colours and sky, and scenery items. Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the place and its item ids. One undo step for the person. For a generated photo background use generate_place; for your own picture use upload with as background. |
| keysA | Animate anything by hand with keyframes on a track. Tracks: "cast:" (x, y, z, size, spin, turn), "item:" (x, z, size), "text:" (x, y, size), "light:" (x, y, z, intensity, color), "scene:light" (level, ambient), "mask:shot", "adjust:", "element:". op set puts values at a time (a first key after 0 also keys the resting value at 0), remove deletes the key at that time, clear drops the track. One undo step for the person. For ready-made moves use act; for the camera use camera keys. |
| transitionA | Set how a shot changes into the next one: a transition id (cut, fade, whip-left, match-zoom-in, ink-splat... see library transitions), its length, and its options. Also the end-of-shot moves before it (outro camera, effect, character), or a ready combo of both. Empty string removes an outro part. Ids must come from the current style: call library to see them (a wrong id is refused with the valid list). Returns the transition. One undo step for the person. |
| list_stylesA | Read-only: every built-in Dinoctor style with its id, name, what it's for and its depth (2D, 2.5D or real 3D), with the project's current style marked. Use an id with library (what it offers), new_project or set_style. The person's own styles are listed by my_styles. |
| libraryA | Read-only: the ids a style offers, by category (characters, objects, places, cameras, actions, emotions, text_presets, text_animations, transitions and outro moves, lights, looks, elements, samples), with names and short notes. With no category and no search it returns counts only; pass category and/or search to keep answers small. Every id you pass to cast, act, words, camera, light, place or transition must come from here. For ready-made shot recipes to add, use samples. |
| how_toA | Read-only reference text: the project format (Project > Scene > Shot fields), units (seconds vs percent), x/z ranges and keyframe tracks. Read it once before using edit_project. It doesn't need the app open. For the drawing format of draw_* tools, use describe_drawing. |
| lookA | See a shot: returns a picture (about 960 px wide) of a shot at a moment, plus its info, so you can check what you made and fix it. Give a shot (id, or scene and shot index; default the shot on screen) and seconds into it, or film: true with seconds into the whole film. It shows that shot in the editor and pauses it; the project isn't changed. Doesn't work while a render runs. For the shot's settings as data use get_shot; for a video file use render. |
| renderA | Start making a video file (MP4) of the film, one scene or one shot, the same way Export does. It returns at once; then call render_status until done (it gives the file path). Only one render runs at a time, and look doesn't work while it runs. quality draft is 540p and quick (for checking timing); final is full size and much slower. make wav or mp3 exports only the sound. Without save_to it goes to Dinoctor's renders folder. To stop it, use cancel_render. |
| render_statusA | Read-only: how the render started by render is going: stage, progress (0 to 1), frames and seconds so far; when done, the file path; or the error. Call every 10 to 30 seconds while running. Returns running: false when no render has run. |
| cancel_renderA | Stop the render that's running (started by render). No video file is made; the project isn't changed. Safe to call when nothing runs. To stop a picture or style job, use picture_status with stop or cancel_style. |
| list_projectsA | Read-only: the person's recent project files (path, name, style, when opened, whether the file still exists), and which file is open now and whether it has unsaved changes. Use a path with open_project. |
| open_projectA | Open a project file (.dino) in Dinoctor, the same as File > Open, in place of the open project. Refuses if the open project has unsaved changes, unless discard: true (ask the person first, or save_project before). Returns the opened project's summary. For a new empty project use new_project; for a template use templates. |
| save_projectA | Save the open project to a .dino file, the same as File > Save. With path: Save As to that file. With name: a new file with that name in the person's projects folder. A project never saved before gets a file in the projects folder. Returns the file path. To save it as a reusable starting point, use templates with save. |
| generator_accountA | Read-only check of the paid services Dinoctor can use and what's left: Higgsfield (pictures: signed in, credits, any job running), ElevenLabs (voices: connected, plan, characters left) and whether Claude Code is here (for writing scripts). Call it before paid tools to see if they will work. sign_in: true opens the person's browser to sign in to Higgsfield themselves. |
| generate_pictureA | Make any picture with Higgsfield (Nano Banana) from words: a character or object (its background is taken off) or a background. Spends the person's Higgsfield credits: the first call (without confirm) only returns the price and credits left; ask the person, then call again with the same arguments and confirm: true. Waits up to about 50 s; if it isn't done, call picture_status. put defaults to media (the Media bin). Use the shortcuts generate_character (straight into a shot's cast) or generate_place (straight in as the shot's background) when that's what you want. For a free drawing in the style itself, use draw_character, draw_object or draw_place. |
| generate_characterA | Make a character picture with Higgsfield (whole body, background taken off) and add it to a shot's cast, or replace a member's picture with replace. Spends the person's Higgsfield credits: the first call (without confirm) only returns the price and credits left; ask the person, then call again with the same arguments and confirm: true. Waits up to about 50 s; if it isn't done, call picture_status. put: media or none keeps it out of the shot. For objects or other kinds use generate_picture; for a free drawn character in the style use draw_character; to reuse one made before use put_picture. |
| generate_placeA | Make a background picture (a place, no people) with Higgsfield and make it the shot's background. Spends the person's Higgsfield credits: the first call (without confirm) only returns the price and credits left; ask the person, then call again with the same arguments and confirm: true. Waits up to about 50 s; if it isn't done, call picture_status. For the style's own places use place with a preset (free); for a free drawn place use draw_place; to reuse one made before use put_picture. |
| picture_statusA | Follow the Higgsfield picture being made (by generate_* or relight): waits up to wait seconds and, when it's done, puts it where that call said and returns the picture id. stop: true stops the job instead. Returns { done: false } with the stage if it's still working, or { job: null } if nothing is being made. |
| generated_picturesA | Read-only list of pictures made with Higgsfield before (kept on this computer for every project), newest first: id, kind, what was asked, when, folder. Free to reuse: pass an id to put_picture, remove_background, relight, or as from_pictures in generate_*. Pictures already in the project are in get_shot instead. |
| put_pictureA | Put a picture made before (an id from generated_pictures) into the project, free: as a character or object (joins a shot's cast, or replaces a member's picture), a background (the shot's place) or media (the Media bin). One undo step. Returns where it went. For a file on this computer use upload; for a new picture use generate_*. |
| remove_backgroundA | Take the background off a picture made before (generated_pictures id) with Higgsfield's remover; the saved picture is changed. Uses a little of the person's Higgsfield credits: the first call only says so; call again with confirm: true after they agree. Optionally puts the result into a shot or the Media bin (as). For free trimming, cropping or flipping of a picture already in a shot, use cutout. |
| relightA | Higgsfield relights a picture photo-real: everything stays, only the light changes (like the Thumbnail studio's Make it real). Give a picture file path or a generated picture id, and the light in words. A new picture is made; the original stays. Spends the person's Higgsfield credits: the first call (without confirm) only returns the price and credits left; ask the person, then call again with the same arguments and confirm: true. Waits up to about 50 s; if it isn't done, call picture_status. The result goes to media, a shot's background, or nowhere. To light a shot in the app itself (free), use light. |
| cutoutA | Free picture fixes on this computer for a picture already in a shot: trim its empty see-through edges, crop a percent off each side, or flip it. Give member (a cast member id with a picture) or leave it out for the shot's background picture. One undo step. Returns the new size. To remove a background (paid), use remove_background. |
| uploadA | Bring a file from this computer into the project, like the app's upload. Pictures (PNG/JPG/WebP/GIF/SVG): as media, character or object (joins a shot's cast or replaces a member's picture), or background. Videos (MP4/MOV/WebM): as media, video-background or video-front in a shot. Sounds (MP3/WAV/M4A...): as sound (sound Media, and on the timeline at |
| get_scriptA | Read-only: the film's script from Sound and script: title, the chosen voice, every line (id, text, feeling, whether it has a voice and how long) and which shots speak it. Use line ids with assign_line, speak or own_voice. To put a script in, use write_script. |
| write_scriptA | Put a script in the film, the way Sound and script does. Give your own lines (short, 5-15 words, one shot each), or text (a whole script, cut into lines at sentences), or only a topic (Dinoctor's Claude writes it; uses the person's own Claude Code and can take a few minutes, so say so first). Unless put_in_film is false, lines go into a scene: shots there without a voice take lines in order, and each line left over becomes a new shot sized to it. Captions show a line's words until it has a voice. replace: false adds to the script. One undo step. Returns the lines and shots. To make the voices, call speak next; to give one shot one line, use assign_line. |
| assign_lineA | Give one shot one script line: the shot speaks that line's voice (or shows its words as a caption until it has one). It replaces the shot's own voice. One undo step. Returns the shot, line and length. To put a whole script into many shots at once, use write_script. |
| voicesA | List the ElevenLabs voices in the person's own account (id, name, what it sounds like), read-only. With pick: a voice id becomes the film's voice for speak (one undo step). Needs ElevenLabs connected (see generator_account). Making voices costs characters: that's speak. |
| speakA | Make the voice for script lines with ElevenLabs, with the time of every word, and give it to the shots that speak each line (captions follow the words; shots fit the voice unless fit is false). Spends the person's ElevenLabs characters: the first call (without confirm) returns the cost and what's left; ask, then call again with confirm: true. Default: every line without a voice, in the film's voice. Returns what was made and characters left. For the person's own recording (free), use own_voice. |
| own_voiceA | Use the person's own recording (a sound file on this computer) as a voice, free: for a script line (line), for a shot (it becomes that shot's line), or as a new line. Its words are heard on this computer when voice text is set up, so captions can follow them. cut_pauses takes out long pauses. One undo step. For an AI voice (paid) use speak; for music use music. |
| musicA | The film's background music: add the person's music file (it's levelled so it sits under the voice; the levelled copy replaces it by itself a minute or two later), set its volume and how far it drops under the voice, or remove it. One undo step. Returns the music settings. For short sounds at a moment, use upload as sound or sound_effects. |
| sound_effectsA | Dinoctor's automatic sound effects (from the style's sound pack, worked out from what happens in each shot). Film-wide: on, volume, room tone. One shot: shot_on, shot_volume, and one sound by its event key: event_on, pick another sound, own_path (the person's own file), gain, nudge, reset. Call with just shot to list that shot's sounds and their event keys (no change). One undo step per change. For music use music; for any other sound file use upload as sound. |
| my_stylesA | Read-only: the person's own art styles (ones Claude built for them in Dinoctor) with id, name, a short about, tokens used and their star rating, plus the style job running now (building or fixing, with progress). Use an id with use_style, fix_style, rate_style or delete_style. Built-in styles are listed by list_styles. |
| make_styleA | Start building a brand-new art style from words and/or example pictures, the same as Dinoctor's "Build your own style". Claude Code runs inside Dinoctor on this computer and builds the style in code: it plans the look, draws characters, objects and places, writes the style, runs the app's tests, checks a test shot and makes the pictures. Usually about 30 minutes (longer for a detailed 3D style), and it uses a lot of the person's own Claude usage (a smaller plan can hit its limit). Nothing is paid to Dinoctor. The person can keep working, and cancel_style stops it at any time. Without confirm: true it only says what will happen; ask the person, then call again with confirm: true. Only one style job runs at a time. Then follow it with style_status. To fix a style that exists use fix_style; to resume a stopped build use continue_style. |
| style_statusA | How the style build or fix (make_style, fix_style, continue_style) is going: status (running, done, failed, cancelled), step and its name, what Claude does right now, minutes so far and left (a rough guess), tokens, errors and problems. Read-only unless close: true, which clears a finished job from Dinoctor's screen. When done, use_style switches to it and rate_style records how it went; when failed with canContinue, use continue_style. To stop it, use cancel_style. |
| continue_styleA | Resume a style build or fix that stopped part way (style_status says canContinue: true), in the same Claude session, from where it stopped. Uses more of the person's Claude usage, so the first call only says what will happen; call again with confirm: true after they agree. Fails if nothing can be continued. To start a new style use make_style; to repair a finished one use fix_style. |
| fix_styleA | Ask Claude to fix a problem in one of the person's own finished styles, like Dinoctor's Fix a problem: it reads the note, fixes the code, runs the tests, checks a test shot and redraws the pictures (about 10 minutes of the person's Claude usage). The style stays as it was until the fix passes. The first call only says what will happen; call again with confirm: true after the person agrees. Follow it with style_status. For a stopped build use continue_style; for a new style use make_style. |
| cancel_styleA | Stop the style build or fix that's running (make_style, fix_style, continue_style). Nothing is changed: the person's styles stay as they were. Returns the job's state, or a note if nothing runs. To only check progress, use style_status. |
| rate_styleA | Save the person's rating of one of their own styles (1 to 5 stars and a comment), as Dinoctor asks after a build or fix. Only rate with what the person told you; never make up a rating. It stays on this computer. Returns { rated, stars }. |
| delete_styleA | Delete one of the person's own styles for good. It can't be undone, and templates in that style stop opening. Without confirm: true it only says what will go; ask the person first. Dinoctor reloads afterwards. Built-in styles can't be deleted. To only stop using it, switch with use_style instead. |
| use_styleA | Switch the project to another art style, a built-in id (list_styles) or one of the person's own (my_styles). Every shot is rebuilt in the new style: its characters, objects and places are replaced; words, cameras and timing stay. One undo step, and no confirm step, so ask the person first if the project has work. A style built since Dinoctor opened makes the window reload once. For built-in styles with a confirm step, set_style does the same. |
| samplesA | Ready-made samples in the current style: whole films and shot runs (on shelves like "Whole films", "Film openings", "Hooks"), shot recipes (groups like "Big scenes", "Instagram hook"), and samples the person saved. action list (read-only; shelf or search narrows it), open (plays one in the preview without adding it; returns its shots and the pictures it needs), close, add (puts its shots in: how add = after the shot, shot = in place of the shot, scene = over the whole scene; returns new shot ids and picture spots to fill), needs (empty picture spots; fill them with fill_spot), save (saves a scene's shots as the person's own sample). Adding is one undo step. For single ids (a camera, an action) use library; for whole saved projects use templates. |
| fill_spotA | Put a picture into an empty picture spot a sample is waiting for (a stand-in person, a photo, a screen or a card's picture). key comes from samples "needs" or "add". Send a picture file from this computer (full path), or the address of a picture Dinoctor already has (starting with /, as generate or upload tools return). One undo step. To put a picture into a shot's cast or background instead, use upload or put_picture. |
| templatesA | The person's templates (saved whole projects to start from), like Dinoctor's Templates page. action list (read-only), save (name: saves the project as it is now), open (id: replaces the open project with the template; one undo step brings it back), delete (id: for good, needs confirm: true after the person agrees). For .dino project files use open_project / save_project; for shot recipes use samples. |
| describe_drawingA | Read-only: the exact drawing format for draw_character, draw_object or draw_place in the current style (or the style given), with the look the style wants and an example when it has one. Read it before drawing. It describes the format, not a shot: to see a shot use look; for the project format use how_to. |
| draw_characterA | Draw a new character yourself, free, in the current style's format, and add it to the person's library and into a shot. Use it for simple characters the style doesn't have; for a photo-like picture use generate_character (paid). No Claude usage or credits are spent: you send the drawing as plain data and Dinoctor checks it (bad fields are dropped; a drawing that fails is refused with the reason). Call describe_drawing first for the exact format. Formats in short: Pixel = {"name","rows":["..##..",...],"colors":{"#":"#535353"},"eyes":[[col,row]],"mouth":[col,row]} (up to 28x28, '.' empty, facing right, at least 20 columns wide); Toon = {"name","skin","shirt","pants","hair":"tuft|bob|ears|antennae|cap|horns|none","hairColor","snout","tail","wide","tall"}; Voxel = {"name","boxes":[{"at":[x,y,z],"size":[w,h,d],"color","part":"body|legL|legR|armL|armR"}],"eyes","mouth","faceZ"}; other styles have their own format (describe_drawing). Places in every style: {"name","sky","horizon","floor","dark","stars","clouds","orb","items":[{"kind","x","z","size"}]}. Returns the new id. Then use look to check it. |
| draw_objectA | Draw a new object yourself, free, in the current style's format, and add it to the person's library and into a shot. Use it for simple objects the style doesn't have; for a photo-like picture use generate_picture (paid). No Claude usage or credits are spent: you send the drawing as plain data and Dinoctor checks it (bad fields are dropped; a drawing that fails is refused with the reason). Call describe_drawing first for the exact format. Formats in short: Pixel = {"name","rows":["..##..",...],"colors":{"#":"#535353"}} (up to 28x28, '.' empty); Toon = {"name","w","h","shapes":[{"type":"rect|ellipse|poly","x","y","w","h","color","points"}]} (y up is negative, stands on y 0); Voxel = {"name","boxes":[{"at":[x,y,z],"size":[w,h,d],"color"}]}; other styles have their own format (describe_drawing). Places in every style: {"name","sky","horizon","floor","dark","stars","clouds","orb","items":[{"kind","x","z","size"}]}. Returns the new id. Then use look to check it. |
| draw_placeA | Draw a new place yourself, free, in the current style's format, and add it to the person's library and set it as the shot's place. Use it for simple places the style doesn't have; for a photo-like picture use generate_place (paid). No Claude usage or credits are spent: you send the drawing as plain data and Dinoctor checks it (bad fields are dropped; a drawing that fails is refused with the reason). Call describe_drawing first for the exact format. Formats in short: Pixel = {"name","rows":["..##..",...],"colors":{"#":"#535353"}} (up to 28x28, '.' empty); Toon = a place object; Voxel = {"name","boxes":[{"at":[x,y,z],"size":[w,h,d],"color"}]}; other styles have their own format (describe_drawing). Places in every style: {"name","sky","horizon","floor","dark","stars","clouds","orb","items":[{"kind","x","z","size"}]}. Returns the new id. Then use look to check it. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 69 tools
Most tools map to distinct resources or actions, and many descriptions explicitly redirect to the right alternative. However, several overlapping families create real ambiguity: use_style vs. set_style, generate_* vs. draw_* vs. upload vs. put_picture, speak vs. own_voice vs. assign_line/write_script, and multiple ways to open content (open_project, templates open, samples open, new_project). The detailed cross-references help, but the 69-tool surface is still hard to navigate.
All names use snake_case, so casing is consistent and readable. But the convention mixes noun-only tools (voices, music, cast, act, words, camera, light, place, keys, transition, library, samples, templates, status, look) with verb-based tools (get_project, add_shot, remove, generate_picture, cancel_render), so there is no predictable verb_noun pattern.
The domain is genuinely broad — project editing, style building, AI picture/voice generation, rendering, audio, scripts, samples and templates — so some breadth is justified. Still, 69 tools is far beyond a typical well-scoped set and likely overwhelms tool selection; it would benefit from grouping or splitting into narrower servers.
The surface covers nearly the full lifecycle: project open/save/list, scene/shot CRUD, casting, acting, text, camera, lighting, places, keyframes, transitions, script and voice handling, sound/music, rendering, style management, generation, uploads, samples and templates. Minor gaps remain, such as no explicit undo/redo tool, no project deletion, and limited deletion/management for media or generated pictures.