ResolveMCP-ext

Lay out a multi-camera comparison timeline (Insta360 grid / split views). Reproduces the camera-comparison workflow this project has been running as loose scripts, with the position bug fixed. **The bug worth knowing:** those scripts set the image position with `SetProperty("PositionX", …)` / `"PositionY"`. MEASURED on 21.1.2.10 those keys return **False and are silently ignored** — `GetProperty` gives back `False` too. The real keys are **`Pan`** (x, +right) and **`Tilt`** (y, +up), in pixels. Zoom worked in those scripts; positioning never did. This tool uses Pan/Tilt and verifies each write by reading it back. Camera detection from the clip name, longest prefix first: ace* -> Insta360 AcePro (marker Orange) go* -> Insta360 Go Ultra (marker Green) luna* -> Insta360 Luna Ultra (marker Cyan) x* -> Insta360 X5 (marker Blue) vid_* -> Handy (marker Gray) (an Android file, e.g. VID_20260614_170228.mp4) Actions: detect(track_type?, track_index?) -> {clips: [{name, id, camera}], unknown} — what this timeline holds and which camera each clip is. Run this first: it tells you whether the names match the convention. grid_2x2(track_type?, track_index?, scale?, verify?) -> {applied, skipped, layout} — place the four cameras in a 2x2 quadrant grid. Quadrant centres are at ±25 % of each axis; default scale 0.5. split_h / split_v(track_type?, track_index?, scale?) -> {applied, layout} — two-up side by side (left/right) or stacked (top/bottom). layout(assignments, track_type?, track_index?, scale?) -> {applied, results} — explicit control. assignments maps a clip_id (or camera name) to {x?, y?, scale?}, where x/y are fractions of the timeline raster (+x right, +y up) and scale multiplies the 1.0 zoom. fullframe(clip_ids?|cameras?, track_type?, track_index?) -> {reset} — back to 1.0 zoom and centred, for the "phone as full picture" clip.…

Copy clip attributes between timeline items (Resolve 21.1.2+). `TimelineItem.CopyAttributes` is the API behind Edit-page Paste Attributes. It is what makes a per-camera look repeatable: grade one clip, then push the same attributes onto every other clip of that camera. Actions: options() -> {groups} — the full option vocabulary, grouped, from Resolve's own type stubs. copy(source_id?, source_name?, target_ids?, target_names?, options, copy_all_video?, copy_all_audio?, track_type?, track_index?) -> {success, copied_to, options_used} — copy attributes from one source item onto one or more targets. `options` is a dict of the flags from options(); a true value means copy that attribute. Shorthands: copy_all_video=true -> allVideoAttributes copy_all_audio=true -> allAudioAttributes Common per-camera case: {"colorCorrection": true, "allZoomAttributes": true, "allCropAttributes": true, "videoPlugins": true} read(item_id?|name?, track_type?, track_index?) -> {attributes} — the readable video attributes of an item (position, zoom, crop, opacity, rotation, composition mode), so you can see what a copy would carry. Notes: - Targets are addressed by `target_ids` (GetUniqueId list) or `target_names` (exact names on one track). Pick one form. - CopyAttributes needs the source and targets on the SAME timeline.

Inspect clip placement, handle offsets, gaps and timeline enable state. **SetPosition / MoveItems are BROKEN on Resolve 21.1.2.10** — measured to return False for every input (real clips and stills, int and float startFrame, even a no-op), while other write APIs work in the same session. `set` and `move` therefore report an honest error rather than pretending to have moved anything. Use `get` to plan, and reposition in the Edit page by hand. Actions that DO work: get(item_id?|name?, track_type?, track_index?) -> {position, left_offset, right_offset, start, end} — where an item sits and how many frames of handle it offers. This is the read you need before deciding a trim. READ-ONLY, safe. enabled() -> {enabled} — read the timeline's enabled state. Setting it (enabled=true/false) works but turns the timeline non-current, so later calls report 'No current timeline' until it is switched back in the GUI. Actions DISABLED or exposing a broken API on this build: set(start_frame?|new_track_index?) — SetPosition returns False for every input, including no-ops. move(items=[…]) — same broken call per item. set_offset(left?|right?) — SetLeftOffset does NOT nudge the edit point: it EXTENDS the clip from source media. Measured: a clip at frame 400 with duration 120 became start=-85995, duration=86515 after SetLeftOffset(5), and could not be restored by setting 0. delete_gaps() — closes the gaps between clips so they sit back to back. **Verified working** on a timeline whose StartFrame/EndFrame agree with its contents. It appeared to be a silent no-op in an earlier release; that measurement ran on a timeline whose metadata contradicted its contents (StartFrame == EndFrame while holding three clips). The action compares the clip starts before and after and reports the real outcome either way. Addressing: `item_id` (GetUniqueId) or `name` on a track…

Clip effects and effect properties on TimelineItems (Resolve 21.1.2+). The upstream server has no coverage for the 21.1.2 "Clip Effect Actions" and "Effect Properties" APIs. This tool exposes them. Actions: list_available(media_type?) -> {effects, count} — every effect Resolve can apply. media_type: 'video' (default), 'photo' or 'audio'. Each entry: {category, id, name}. This is the lookup you need before apply(). Example: find the id for a Fusion text+ via list_available(media_type="video"). list_applied(item_id?|name?, track_type?, track_index?, category?) -> {effects, count} — effects currently ON a timeline item. category filters to 'fusion' | 'ofx' | 'audio'; omit for all three. apply(item_id?|name?, category, id, track_type?, track_index?) -> {effect: {category,id,name,index,enabled}} — applies an effect to a timeline item. Get category+id from list_available(). remove(item_id?|name?, indices?|category?, track_type?, track_index?) -> {removed: [...]} — removes effects from a timeline item. Either give explicit 1-based `indices` (with `category`), or `category` alone to remove every effect of that category. set_enabled(item_id?|name?, index, enabled, category?, track_type?, track_index?) -> {success, effect} — toggle one effect on/off by its index. reorder(item_id?|name?, index, new_index, category?, track_type?, track_index?) -> {success, effects} — move an effect to another position in its category's stack. Addressing a timeline item: give `item_id` (GetUniqueId) — or `name` plus the track it sits on. Track defaults are video/1.

Export a frame as a still — from a timeline, a clip, or the media pool. Resolve 21.1.2 added ExportFrame on Timeline, TimelineItem and MediaPoolItem. The upstream server's frame capture goes through a render or a Gallery still; this is the direct call. Actions: item(item_id?|name?, frame_offset?, output_type?, file_path?, format?, max_dimension?, track_type?, track_index?) — TimelineItem.ExportFrame. **This is the one that works.** Offset is 0-based from the ITEM start and must be < the item's duration (verified: 0..119 succeed on a 120-frame item, 120+ is refused). media(clip_id, frame_offset?, output_type?, file_path?, format?, max_dimension?) — MediaPoolItem.ExportFrame. Also verified working; offset is 0-based from the SOURCE start, ignoring the timeline. timeline(frame_offset, …) — Timeline.ExportFrame. Offset is **0-based from the TIMELINE start** (`StartFrame`), valid `0 .. (EndFrame - StartFrame - 1)`. Verified on 21.1.2.10 against a 480-frame timeline: offsets 0..479 succeed, 480 is refused with 'frameOffset is out of range'. An absolute frame (e.g. the value from `GetStartFrame()`) is ALWAYS refused, which reads like a failure but is correct behaviour. An earlier release documented that rejection as a defect; that was a wrong measurement on an empty timeline, corrected here. Parameters: output_type 'file' (needs file_path) or 'encodedString' (default). file_path Output path; format is inferred from the extension. format 'jpg' (default) or 'png' — only for output_type 'encodedString'. max_dimension Cap the longer edge in pixels (default 0 = full size). Returns for output_type='file': {success, path, bytes}. Returns for 'encodedString': {saved_to} — the base64 is decoded and written to <server dir>/out/, with only a short prefix returned inline.

Folder hierarchy lookups the upstream server does not expose. Actions: parent_of_clip(clip_id) -> {name, id, path} — which bin a media pool clip lives in (MediaPoolItem.GetParentFolder). parent_of_folder(path?|folder_id?) -> {name, id, path} | {root: true} — a folder's parent (Folder.GetParentFolder). The root folder has none. path(clip_id?|folder_id?) -> {path, segments} — the full "Master/Sub/Bin" path of a clip or folder. find_clip(clip_id) -> {name, folder_path} — locate a clip by id anywhere in the tree.

Node graph inspection and labelling, on a clip or a photo (21.1.2+). Verified working on 21.1.2.10: GetNodeLabel, SetNodeLabel, GetNodeColor, SetNodeColor, SetNodeEnabled, GetNumNodes, ResetAllGrades, ResetAllNodeColors. SetLUT/GetLUT returned False/None in testing (a LUT path is required and must exist) — the tool reports what it gets rather than assuming. Target: pass `item_id`/`name` for a timeline clip, or `photo_id` for a photo album item. `clip_id` is accepted but NOT usable on 21.1.2.10: MediaPoolItem.GetNodeGraph is not callable there, while the timeline and photo variants both work. Actions: info(item_id?|name?|photo_id?|clip_id?, track_type?, track_index?) -> {num_nodes, nodes: [{index, label, color, enabled}]} label(node_index, label, ...) -> {success, nodes} — SetNodeLabel. Pass an empty string to clear. color(node_index, color, ...) -> {success, nodes} — SetNodeColor. Valid names: Orange, Apricot, Yellow, Lime, Olive, Green, Teal, Navy, Blue, Purple, Violet, Pink, Tan, Beige, Brown, Chocolate. An empty string clears the color. enable(node_index, enabled, ...) -> {success, nodes} tools(node_index, ...) -> {tools} — which Fusion/OFX tools sit in that node. lut(node_index, lut_path?, ...) -> {success, lut} — read, or set when lut_path is given (Resolve's relative LUT path). cache(node_index, mode?, ...) -> {success, cache_mode} reset_colors(...) -> {success} reset_grades(...) -> {success} — destructive: clears the whole grade.

Photo albums and their items (Resolve 21.1.2+). A photo album is the stills/photo counterpart of a timeline: an ordered set of photo items, each with its own grade, flags, rating and versions. Measured on 21.1.2.10 this whole area works — 9/9 PhotoAlbum methods and 20/20 PhotoAlbumItem methods answered. The API is split across objects and the split is NOT documented: the album *listing* methods live on Project, while creating/adding lives on MediaPool, and Resolve itself answers none of them. This tool resolves the right object at call time, so it keeps working if a build moves them. Albums: list() -> {albums, count} current() -> {album|None} set_current(album_id?|name?) -> {success, album} create(name) -> {album} create_from_photos(name, clip_ids) -> {album} — THE RELIABLE WAY to get photos into an album: VERIFIED working (added 2 photos, read back). Prefer this over add_photos. delete(album_ids?) -> {success, deleted} rename(album_id?|name?, new_name) -> {success, album} quick_export_presets() -> {presets} render_settings(settings?) -> {settings} Photos inside an album: photos(album_id?|name?) -> {photos, count} add_photos(clip_ids, album_id?|name?) -> {added, count} — BROKEN on 21.1.2.10: AddToPhotoAlbum adds nothing and returns an empty list. This action verifies by reading the album back and reports the failure with the measurement instead of a false success. Use create_from_photos instead (verified working). remove_photos(photo_ids, album_id?|name?) -> {success, remaining} current_item(album_id?|name?) -> {photo|None} Per-photo properties (address a photo by photo_id, or by name inside an album): photo(photo_id|photo_name?, album_id?|name?) -> {photo} rate(photo_id|photo_name?, stars) -> {success, rating} flag(photo_id|photo_name?, color?|clear?) -> {success, flags} clip_color(photo_id|photo_name?, color?|clear?) -> {success, clip_color} versions(photo_id|photo_name?) -> {versions, curre…

Project-level gaps: render presets, audio render formats, transcription. Three surfaces the mainstream servers do not implement, all verified present on Resolve 21.1.2.10 by calling them (the type stubs list them, but a stub only says a name exists — see CONTRIBUTING.md). Render presets: render_presets() -> {presets} — GetRenderPresetList. Verified: returns the real list. update_render_preset(preset_name) -> {success, preset} — UpdateRenderPreset: overwrite a saved preset with the CURRENT render settings. MEASURED on 21.1.2.10: returns False for the 37 factory presets GetRenderPresetList reports (e.g. 'H.264 Master'), so this only has a chance against a preset saved from the Render page in the GUI. Project.SaveRenderPreset does NOT exist in the API, so a preset cannot be created from here to test with. Treat as GUI-dependent. quick_export_enable(preset_name, enabled) -> {success} — SetQuickExportEnabledForRenderPreset. Same limitation: False for factory presets on this build. Audio render formats: audio_formats() -> {formats} — GetAudioRenderFormats: {display name -> file extension}. audio_codecs(format_extension) -> {codecs} — GetAudioRenderCodecs: {codec description -> codec name} for a format extension such as 'wav', 'mov', 'mxf'. Transcription (a clip's speech-to-text content): transcription(clip_id, use_nested?) -> {transcription} — GetTranscription. `use_nested` reads a nested timeline's transcript. set_transcription(clip_id, transcription) -> {success, verified} — SetTranscription. REPLACES an existing transcript. MEASURED on 21.1.2.10: returns False for a PNG still AND for a real video clip that had no transcript, with both '01:00:00:00' and '00:00:00:00' start timecodes — it appears unable to create a transcript where none exists. Produce one in the Edit page (AI Transcription) first, then this can replace it. The shape is: {languag…

Health check for the resolve-ext server and its Resolve connection. Reports whether Resolve is reachable, which product/version is live, whether a project is open, and which of this server's surfaces are usable on that build.