| name | davinci-resolve-mcp |
| description | Control DaVinci Resolve via the davinci-resolve MCP server. Use this skill whenever the user asks to work with DaVinci Resolve, video editing, color grading, timelines, media pools, rendering, Fusion compositions, or any Resolve-related task. Triggers include: mentions of "Resolve", "DaVinci", "timeline", "color grade", "render", "media pool", "Fusion comp", "captions", "subtitles", or any video post-production workflow. Covers 440+ MCP tools spanning the full Resolve scripting API and Fusion composition API.
|
DaVinci Resolve MCP Server
Control DaVinci Resolve programmatically through 440+ MCP tools.
AI Assistant <--> MCP Protocol <--> davinci-resolve-mcp <--> fusionscript <--> DaVinci Resolve
Critical Concepts
Object Registry
Resolve API objects cannot be serialized. The server maps string IDs to live objects.
Always follow this chain:
- Get a parent object (returns an ID)
- Pass that ID to child tools
- If objects go stale, call
clear_object_cache and re-fetch
resolve_get_project_manager --> project_manager_id
--> project_manager_get_current_project --> project_id
--> project_get_current_timeline --> timeline_id
--> timeline_get_item_list_in_track --> [timeline_item_ids]
Response Format
All tools return JSON: {"success": true, "result": ...} or {"success": false, "error": "..."}.
Page Requirements
Some tools require specific Resolve pages:
- Thumbnails: Color page
- Fairlight audio: Fairlight page
- Fusion comps: Fusion page (or via timeline item bridge)
- Stills/Gallery: Color page
Use resolve_open_page("color") etc. to switch first.
Tool Categories
| Category | Count | Key Tools |
|---|
| Resolve | 22 | resolve_open_page, resolve_get_version, layout/render presets |
| ProjectManager | 25 | project_manager_get_current_project, project_manager_load_project, database/cloud |
| Project | 43 | project_get_current_timeline, project_set_render_settings, project_start_rendering |
| MediaStorage | 7 | media_storage_get_file_list, media_storage_add_item_list_to_media_pool |
| MediaPool | 27 | media_pool_import_media, media_pool_create_empty_timeline, media_pool_append_to_timeline |
| Folder | 8 | folder_get_clip_list, folder_get_sub_folder_list |
| MediaPoolItem | 36 | metadata, markers, flags, clip properties, proxy, transcription |
| Timeline | 59 | tracks, markers, titles, timeline_create_subtitles_from_audio, export, scene detection |
| TimelineItem | 80 | properties, Fusion comps, color versions, takes, CDL, grades, stabilization |
| Gallery | 14 | still/PowerGrade albums, import/export stills |
| Graph | 11 | node LUTs, cache modes, grade application |
| ColorGroup | 5 | naming, pre/post clip node graphs |
| Utility | 3 | clear_object_cache, initialize |
| Fusion Comp | 40 | comp_add_tool, comp_find_tool, undo/redo, render, lock/unlock |
| Fusion Tool | 20 | tool_set_input, tool_connect_input, modifiers, settings |
| Fusion Input | 12 | expressions, connections, keyframes |
Common Workflows
See references/workflows.md for step-by-step recipes covering:
- Getting started (project/timeline chain)
- Importing media and building timelines
- Setting up and running renders
- Color grading with node graphs
- Batch operations on clips
See references/ffmpeg_workflows.md for audio-driven editing:
- Silence removal (auto jump-cut)
- Audio fingerprint matching (detect game sounds, jingles, effects)
- Volume spike detection (find hype moments)
Fusion Reference
See references/fusion_tools.md for Fusion node types, common inputs (TextPlus, Merge, Background), and connection patterns.
FFmpeg + Resolve Integration
FFmpeg is used for audio analysis BEFORE building timelines in Resolve. The pattern is:
- Get the source file path via
media_pool_item_get_clip_property(clip_id, "File Path")
- Run ffmpeg analysis (silence detection, audio fingerprinting) on the source file
- Parse results into frame-based segments
- Build timelines using
media_pool_create_empty_timeline + media_pool_append_to_timeline with clip_infos
Silence Removal
ffmpeg -t <duration> -i <file> -map 0:<audio_stream> -af silencedetect=noise=-30dB:d=1.0 -vn -f null -
- Use
-map 0:<N> to target a specific audio stream (MKV files often have multiple)
- Use
-t <seconds> to limit to clip duration (source files may be longer than timeline clips)
- Parse silence_start/silence_end pairs, compute speech segments as complement
- Filter speech segments by minimum duration (0.3s) to avoid micro-clips
Audio Fingerprint Matching
Use FFT-based normalized cross-correlation (see find_sounds_fast.py pattern):
- Extract target audio stream as mono 16kHz WAV:
ffmpeg -i <file> -map 0:<N> -ac 1 -ar 16000 <output.wav>
- Convert reference samples to same format
- Run numpy FFT cross-correlation with normalized scoring
- Filter by confidence threshold (0.7+ recommended for game sounds with background music)
- Merge overlapping ±padding windows
- Build timeline from merged segments chronologically
Performance: ~78 seconds to scan 2.5 hours of audio against 4 sound samples at 16kHz mono.
Timeline Assembly from Segments
media_pool_append_to_timeline with clip_infos is the workhorse:
{"media_pool_item_id": "<id>", "startFrame": 1000, "endFrame": 2000}
- Batch up to 50 clip_infos per call for reliability
- Source frames are 0-based from the media file start
- After MCP reconnection, re-fetch media pool items via
media_pool_get_root_folder → folder_get_clip_list
Important Notes
- Get the project/timeline chain before operating on items
- Use
clear_object_cache when switching projects via the Resolve UI
- Use
initialize if Resolve was restarted after the MCP server started
- MKV files often have multiple audio streams; check with
ffmpeg -i <file> and use -map 0:<N> to target the right one
- Get source time ranges via
timeline_item_get_source_start_time / timeline_item_get_source_end_time
- After MCP server reconnection, object cache is cleared — re-fetch objects before using them
- Wrap batch Fusion operations in
comp_lock/comp_unlock
- Wrap undoable Fusion changes in
comp_start_undo/comp_end_undo
- Many Resolve methods return False/None on failure - always check the success field
