Resolve AI Toolkit

March 16, 2026 · View on GitHub

AI-powered MCP toolkit for DaVinci Resolve Studio — 21 tools across Fusion/VFX, editing, audio, color grading, and organization.

Built for editors who know their craft and want AI to handle the repetitive parts: Fusion comp building, pacing analysis, metadata organization, look development, and audio management — all through natural language via Claude.

What This Does

This is an MCP (Model Context Protocol) server that gives Claude direct access to DaVinci Resolve's Scripting API. Instead of clicking through menus, you describe what you want:

"Apply a warm golden hour look to the hero shot on V2"
"Analyze the pacing of my edit — is it too slow for a commercial?"
"Scan all my clips and organize them by camera"
"Build a lower third that says 'John Smith, CEO'"
"List all Fairlight presets and apply Vanguard_Nick"

Claude translates these into precise API calls through 21 specialized tools.

Requirements

DaVinci Resolve Studio (free version doesn't support external scripting)
macOS (Apple Silicon or Intel)
Python 3.10–3.12 (Resolve's scripting API doesn't support 3.13+)
External scripting enabled in Resolve: Preferences → General → External scripting → Local

Installation

1. Clone the repo

git clone https://github.com/InfiniviewStudios/resolve-ai-toolkit.git
cd resolve-ai-toolkit

2. Create a virtual environment and install

python3 -m venv .venv
source .venv/bin/activate
pip install -e .

3. Verify it works

python -c "from resolve_toolkit.server import mcp, VERSION; print(f'v{VERSION} — {mcp.name} ready')"

4. Configure Claude Code

Add to ~/.claude/settings.json under mcpServers:

{
  "mcpServers": {
    "resolve-ai-toolkit": {
      "command": "/path/to/resolve-ai-toolkit/.venv/bin/python",
      "args": ["-m", "resolve_toolkit"]
    }
  }
}

Replace /path/to/ with the actual path to your clone.

5. Start using it

Launch DaVinci Resolve Studio, open a project, then talk to Claude:

"Analyze my current timeline"
"What does the grade look like on V2 clip 0?"
"Insert a Fusion generator called Text+"

Tool Reference

Every tool uses an action dispatch pattern:

tool_name(action="action_name", params={"key": "value"})

Most tools default to track_type="video", track_index=1, item_index=0 when addressing timeline items.

Fusion/VFX Module (5 tools)

The primary module — automates Fusion composition work that normally requires manual node wiring.

`fusion_comp` — Composition Management

Manage Fusion compositions on timeline items.

Action	Parameters	Returns
`list_comps`	track_index?, item_index?	List of comp names and count
`inspect`	comp_index?, track_index?, item_index?	All tools with IDs and key inputs
`create`	track_index?, item_index?	New empty comp name
`import_comp`	path, track_index?, item_index?	Success status
`export_comp`	path, comp_index?, track_index?, item_index?	Success status
`delete`	name, track_index?, item_index?	Success status

Example: Inspect what's inside a Fusion comp on V5:

fusion_comp(action="inspect", params={"track_index": 5, "item_index": 0})

`fusion_tools` — Node/Tool Manipulation

Read and write any input on any Fusion tool — the power tool for VFX work.

Action	Parameters	Returns
`list_tools`	comp_index?, track_index?, item_index?	All tools with name, ID, passthrough state
`get_input`	tool_name, input_name	Current value
`set_input`	tool_name, input_name, value	Success + readback
`get_all_inputs`	tool_name	Full input dump
`set_passthrough`	tool_name, enabled	Success status

Example: Change a Background node's color to deep blue:

fusion_tools(action="set_input", params={
    "tool_name": "Background1",
    "input_name": "TopLeftBlue",
    "value": 0.8,
    "track_index": 5
})

`fusion_text` — TextPlus Operations

Specialized tools for reading and writing text on Fusion generators.

Action	Parameters	Returns
`scan_text`	min_track? (default 5)	All generators with text field values
`set_text`	track_index, item_index, tool_name, field, value	Success + readback
`set_font`	track_index, item_index, tool_name, font, style?, size?	Success
`set_color`	track_index, item_index, tool_name, r, g, b, a?	Success

Example: Scan all text generators on tracks V5 and above:

fusion_text(action="scan_text", params={"min_track": 5})

`fusion_effects` — Effect Library

Insert Fusion generators, titles, and blank compositions into the timeline.

Action	Parameters	Returns
`insert_generator`	name	Success status
`insert_title`	name	Success status
`insert_composition`	—	Success status
`list_generators`	—	Common generator/title names

Example: Insert a Text+ generator at the playhead:

fusion_effects(action="insert_generator", params={"name": "Text+"})

`fusion_recipes` — AI-Guided Comp Building

Build complete Fusion compositions from recipes — no manual node wiring needed.

Action	Parameters	Returns
`list_recipes`	—	All 10 recipes with descriptions
`build`	recipe, + recipe params	Tools created, tool names
`build_lower_third`	text, subtitle?, font?, size?, color?	Shortcut for lower third
`build_title_card`	text, font?, size?, color?, bg_color?	Shortcut for title card

Available recipes:

Recipe	Category	What it builds
`lower_third`	Titles	Background bar + name TextPlus + optional subtitle
`title_card`	Titles	Full-screen background + centered title text
`name_super`	Titles	Simple name with optional title line
`text_overlay`	Titles	Positioned text (center, top, bottom, corners)
`chroma_key`	Compositing	DeltaKeyer configured for green/blue screen
`luma_key`	Compositing	Luminance-based key
`blend_layers`	Compositing	Merge with blend mode control
`solid_background`	Motion Graphics	Solid color or gradient background
`vignette`	Motion Graphics	Elliptical mask vignette with strength/softness
`blur_transition`	Motion Graphics	Directional blur transition

Example: Build a lower third on V5 clip 0:

fusion_recipes(action="build", params={
    "recipe": "lower_third",
    "text": "John Smith",
    "subtitle": "Chief Executive Officer",
    "track_index": 5,
    "item_index": 0
})

Audio/Fairlight Module (4 tools)

Manage audio tracks, apply Fairlight presets, handle transcription, and analyze audio placement.

`audio_manage` — Track Management

Action	Parameters	Returns
`list_tracks`	—	All audio tracks with names, item counts, lock/enable state
`insert_audio`	path, track_index?	Success status
`get_voice_isolation`	track_index?, item_index?	Current state
`set_voice_isolation`	enabled, track_index?, item_index?	Success status
`add_track`	sub_type? (mono/stereo/5.1/7.1/adaptive)	Success status

`audio_presets` — Fairlight Presets

Action	Parameters	Returns
`list_presets`	—	Available Fairlight presets
`apply_preset`	name	Success status
`scan_audio_tracks`	—	Per-track: clip count, total duration, empty status

`audio_transcribe` — Transcription & Subtitles

Action	Parameters	Returns
`create_subtitles`	options?	Triggers Resolve's built-in subtitle generation
`list_subtitle_tracks`	—	Subtitle tracks with item counts
`export_subtitles`	track_index?	All subtitle items with start/end/duration

`audio_analyze` — Analysis & Placement

Action	Parameters	Returns
`find_gaps`	track_index?, min_gap_frames?	Empty gaps in audio tracks
`coverage_report`	—	Per-track audio coverage percentage
`suggest_music_placement`	edit_track?	Suggested music cue points based on edit rhythm

Example: Get a coverage report across all audio tracks:

audio_analyze(action="coverage_report")

Editing Module (5 tools)

Timeline analysis, clip management, pacing analysis with style-based trim suggestions, marker management, and assembly helpers.

`edit_timeline` — Timeline Operations

Action	Parameters	Returns
`analyze`	—	Full analysis: name, duration, all tracks, item counts, markers
`detect_scene_cuts`	—	Scene cut detection results
`duplicate`	name?	Safe duplicate of current timeline
`export`	path, format? (AAF/EDL/FCPXML/DRT)	Export file

Example: Get a complete picture of your timeline:

edit_timeline(action="analyze")

`edit_clips` — Clip Operations

Action	Parameters	Returns
`list_clips`	track_type?, track_index?	All clips with name, duration, start/end, source info
`get_properties`	track_index?, item_index?	Full property dump (Pan, Zoom, Rotation, Opacity, etc.)
`set_property`	track_index?, item_index?, key, value	Set any clip property
`create_compound`	track_index?, clip_indices?	Create compound clip
`create_fusion_clip`	track_index?, clip_indices?	Create Fusion clip

`edit_pacing` — Pacing Analysis

Analyzes your edit rhythm and suggests trims for different content styles.

Action	Parameters	Returns
`analyze`	track_index?	Shot durations, rhythm classification, energy, tempo, outliers
`suggest_trims`	style?, track_index?	Per-clip trim suggestions with priority
`analyze_transitions`	track_index?	Transition types and durations

Style targets:

Style	Avg Shot	Max Shot	Best for
`documentary`	6.0s	20.0s	Interview-driven docs
`commercial`	2.5s	6.0s	TV spots, ads
`narrative`	5.0s	15.0s	Film, drama, story
`music_video`	1.8s	4.0s	Music videos, montage
`corporate`	4.0s	10.0s	Corporate videos, training
`social_media`	1.5s	3.0s	Reels, TikTok, shorts

Example: Are my shots too long for a commercial?

edit_pacing(action="suggest_trims", params={"style": "commercial", "track_index": 2})

`edit_markers` — Marker Management

Action	Parameters	Returns
`list`	scope? (timeline/clip)	All markers with frame, color, name, note
`add`	frame, color, name?, note?, duration?	Add single marker
`delete`	frame	Delete marker at frame
`bulk_add`	markers (list of {frame, color, name?, note?})	Batch add markers
`export`	—	Export all markers as JSON

Available marker colors: Blue, Cyan, Green, Yellow, Red, Pink, Purple, Fuchsia, Rose, Lavender, Sky, Mint, Lemon, Sand, Cocoa, Cream.

`edit_assembly` — Rough Cut Helpers

Action	Parameters	Returns
`build_selects`	color?, marker_color?, output_name?	Find clips matching criteria for selects timeline
`reorder_by_timecode`	track_index?	Report clips in source timecode order
`find_gaps`	track_type?, track_index?, min_gap_frames?	Find empty gaps between clips

Color Module (4 tools)

Natural-language color grading with 30 built-in cinematic presets.

`look_develop` — Look Development

Action	Parameters	Returns
`generate`	description	CDL values + human-readable description
`apply`	cdl, track_index?, item_index?	Apply CDL to clip
`list_presets`	category?	All 30 presets with descriptions
`apply_preset`	name, track_index?, item_index?	Apply named preset

30 presets in 5 categories:

Film stocks: kodak_5219, fuji_3513, kodachrome, ektachrome, portra_400, cinestill_800t
Moods: golden_hour, blue_hour, moonlight, overcast, harsh_midday, sunset_blaze
Genres: film_noir, scifi_cool, horror_desat, romance_warm, western_dust, thriller_tense
Decades: seventies_warm, eighties_neon, nineties_grunge, teal_and_orange
Emotions: melancholy, joy, tension, nostalgia, isolation, dreamlike, fury, serenity

`shot_match` — Shot Matching

Action	Parameters	Returns
`analyze_hero`	track_index?, item_index?	Hero shot CDL + description
`copy_to_all`	track_index?, item_index?	Copy hero grade to all other clips
`apply_cdl_to_track`	cdl, track_index?	Apply CDL to every clip on a track

`grade_analyze` — Grade Analysis

Action	Parameters	Returns
`describe`	track_index?, item_index?	Plain-language grade description + node info
`compare`	item_a, item_b	Distance score + match quality
`scan_track`	track_index?	Describe every grade on a track

`mood_grade` — Emotional Color Grading

Action	Parameters	Returns
`translate`	description	CDL + dimension scores + confidence
`apply`	description, track_index?, item_index?	Translate and apply
`blend`	description_a, description_b, mix?	Blend two moods
`reset`	track_index?, item_index?	Reset to neutral CDL

Example: Apply "fading memory" mood to V2:

mood_grade(action="apply", params={"description": "fading memory", "track_index": 2})

Organization Module (3 tools)

Smart clip grouping, Media Pool bin management, and batch metadata operations.

`organize_clips` — Smart Grouping

Action	Parameters	Returns
`scan_metadata`	folder?	All clip metadata: cameras, codecs, properties
`group_by`	criterion	Groups by camera, scene, codec, resolution, clip_color
`auto_color_code`	criterion?	Assign Resolve clip colors to groups automatically

`organize_bins` — Bin Management

Action	Parameters	Returns
`list_folders`	—	Top-level Media Pool folders
`create_folder`	name	Create new folder
`move_clips`	clip_ids, target_folder	Move clips to folder
`auto_organize`	criterion?	Create folders by criterion and sort clips

`organize_metadata` — Metadata Ops

Action	Parameters	Returns
`batch_read`	clip_ids?, fields?	Read metadata from clips
`batch_write`	updates [{clip_id, field, value}]	Write metadata in batch
`copy_metadata`	source_id, target_ids, fields?	Copy metadata between clips

Example: Organize all clips by camera and color-code them:

organize_clips(action="auto_color_code", params={"criterion": "camera"})

Workflow Examples

Quick Corporate Video Setup

1. "Analyze my timeline"
2. "Scan the audio tracks — which ones have content?"
3. "Check the pacing for a corporate style"
4. "Apply the 'kodak_5219' preset to V2 clip 0"
5. "Copy that grade to all clips on V2"

Fusion VFX Session

1. "List the Fusion comps on V5 clip 0"
2. "Inspect the comp — what tools are in it?"
3. "Set the StyledText on TextPlus1 to 'New Title'"
4. "Build a lower third on V6 clip 0: text 'Jane Doe', subtitle 'Director'"
5. "Build a vignette on V2 clip 0 with strength 0.3"

Audio Post Session

1. "List all Fairlight presets"
2. "Apply preset 'Dialogue Clean' to the timeline"
3. "Create subtitles from the audio"
4. "Find gaps in audio track 1 longer than 24 frames"
5. "Suggest where to place music cues based on the edit"

Color Grading Session

1. "Generate a look for 'warm sunset with slightly crushed shadows'"
2. "Apply it to V2 clip 0"
3. "Describe what the grade looks like on clip 3"
4. "Compare clip 0 and clip 3 grades"
5. "Blend 'golden hour' with 'film noir' at 30% mix"

Architecture

resolve_toolkit/
├── server.py          # Main MCP server — registers all 21 tools
├── connection.py      # Shared Resolve connection (lazy) + helpers
├── fusion/
│   ├── tools.py       # 5 tools: comp, tools, text, effects, recipes
│   └── templates.py   # 10 recipe builders
├── audio/
│   └── tools.py       # 4 tools: manage, presets, transcribe, analyze
├── editing/
│   ├── tools.py       # 5 tools: timeline, clips, pacing, markers, assembly
│   └── pacing.py      # Pacing analysis engine
├── color/
│   ├── tools.py       # 4 tools: look_develop, shot_match, grade_analyze, mood_grade
│   ├── engine.py      # CDL math, mood-to-CDL translation, describe_cdl
│   └── presets.py     # 30 cinematic look presets as CDL values
└── organize/
    └── tools.py       # 3 tools: clips, bins, metadata

Key Design Decisions

Single MCP server — one process, all 21 tools, no multi-server complexity
Lazy Resolve connection — server starts instantly, connects on first tool call
Action dispatch — each tool handles multiple related actions via action parameter
Shared connection module — all tools import from connection.py for DRY Resolve access
No external AI dependencies — mood-to-CDL translation and pacing analysis are pure Python engines, no LLM calls
macOS-native — Resolve scripting paths hardcoded for macOS (where Resolve Studio runs)

Resolve API Notes

Things to know when working with the tools:

GetItemListInTrack() returns None for empty tracks, not [] — all tools handle this
Fusion comp indices are 1-based — GetFusionCompByIndex(1) is the first comp
GetLeftOffset() returns None for generators and transitions — used as a type detector
CopyGrades() requires Color page active — the shot_match tool uses this
Timeline handles go stale after SetCurrentTimeline() — tools reacquire as needed
Fusion input writes need settling time — tools include time.sleep(0.2-0.3) after writes

License

MIT