justhung69justhung69/resolve-mcp/skills/davinci-resolve/SKILL.md

davinci-resolve

Skill1 starsChanged 31 days agoMITView on GitHub

Use when the user asks to edit, cut, color grade, keyframe, fix audio in, render or inspect a project in DaVinci Resolve - e.g. "cut clip 3 at 5 seconds", "add a cross dissolve between every clip", "brighten the subject with a circle window", "render this timeline for YouTube", "what changed in my Resolve project". Also use when the user mentions Resolve, DaVinci, a .drp project or the Edit/Color/Fusion/Fairlight/Deliver page. Drives DaVinci Resolve 21.0.4 Free on Windows through the resolve-mcp MCP server and verifies every change.

5 other projects use a version of this.

Adopted by

Most recently taken up by

LevyBytes/AI-SKILL-davinci-resolvetheir own version, first seen Oct 23, 2026
matthull/my-skillstheir own version, first seen Oct 20, 2026
odaiouattara0022-del/davinci-resolve-skilltheir own version, first seen Oct 9, 2026
jablonkai/skillstheir own version, first seen Oct 1, 2026
maxenko/claude-skillstheir own version, first seen Jul 14, 2026

Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
jablonkai/skills★ 2same name
maxenko/claude-skills★ 1same name
odaiouattara0022-del/davinci-resolve-skill★ 1same name
LevyBytes/AI-SKILL-davinci-resolve★ 0same name
matthull/my-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

9 rules. 0 are common, 2 are distinctive, and 7 are very specific to this project.

0
Common
Most of the 4 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Very specific:
If it reports bridge: not running, ask the user to open a project in Resolve and click
line 18
Very specific:
If no resolve-mcp tools exist at all, the server is not installed.
line 21
Very specific:
Prefer dry_run=true first for batch edits and show the user the plan.
line 60
Very specific:
Do not enable python_exec.
line 65
Very specific:
Look first: frame_preview before and after.
line 68
Very specific:
Place windows with center_x / center_y (frame fractions 0–1, y down) rather than pan/tilt: one pan unit ≈ 82 px, so pan 65 is already …
line 72
Very specific:
Only circle windows and primaries are render-verified.
line 75
Distinctive:
After each task, reply in this shape (short, in the user's language):
line 91
Distinctive:
If something could not be done, say which limit from section 5 applies and offer the UI steps.
line 99

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.

Files

SKILL.md

DaVinci Resolve via resolve-mcp

This skill drives DaVinci Resolve through the resolve-mcp MCP server (https://github.com/justhung69/resolve-mcp). The server talks to a bridge script running inside Resolve. It works on Resolve 21.0.4 Free on Windows. Resolve 21.1+ Free removed Python scripting, so never tell the user to update Resolve.

Tool names below are the server's tool names. Your client may show them with a prefix, for example mcp__DaVinci_Resolve__Free___resolve_status.

0. Before anything: is the bridge up?

Call resolve_status.
If it reports bridge: not running, ask the user to open a project in Resolve and click Workspace → Scripts → ClaudeBridge. This is needed once per Resolve launch; there is no autostart. Then call resolve_status again.
If no resolve-mcp tools exist at all, the server is not installed. Point the user to the README: run install.ps1, then install the .mcpb Desktop Extension from the latest release.

1. The working loop (always)

Snapshot: get_full_state (depth 1–2 is enough for most tasks). Keep the returned snapshot id.
Act with the most specific typed tool (table below). Use api_call only for documented methods that no typed tool covers. api_catalog lists what exists. Never invent method names.
Check every result's ok. Writes return {requested, actual, ok}. ok: false means Resolve ignored the change even if it returned True. Read notes, fix the input, or explain the limitation. verified: null means Resolve offers no getter for that value. Say so rather than claiming success.
Confirm: diff_state(before_id) shows exactly what changed. For anything visual, call frame_preview and look at the image before telling the user it worked.

2. Which tool for which job

Task	Tools
Project: list/open/create/save/export .drp	`project_list`, `project_open`, `project_create`, `project_save`, `project_export`
Import media, browse bins, clip metadata	`media_import`, `media_pool_list`, `media_folder`, `clip_inspect`, `clip_update`, `media_storage_browse`
Timelines: create, append, insert, tracks, playhead	`timeline_create`, `timeline_append`, `timeline_insert`, `timeline_track`, `timeline_playhead`, `timeline_inspect`, `timeline_list`, `timeline_set_current`
Duplicate, rename, delete timeline; grab still; scene cuts	`timeline_action`
Markers	`markers`
Inspector values (zoom, position, crop, opacity, composite)	`item_inspect`, `item_set_properties`, `item_update`
Cut / razor	`timeline_razor`. Builds a NEW timeline, the original is untouched
Trim, ripple, slip, slide, reorder, gaps	`edit_list_get` → edit the list → `edit_list_build` (new timeline)
Transitions, speed changes	`timeline_build_otio` (new timeline)
Color: LUT, CDL, DRX still, versions, groups, copy grade	`color_apply`, `color_versions`, `color_groups`, `color_copy_grade`, `color_node_graph`, `gallery`
Color: wheels, curves, hue/sat, qualifiers as a LUT	`color_grade_lut`
Color: real Color-page grade with a circle power window	`color_grade_drx`
Keyframed zoom / position / rotation / opacity	`keyframe_transform` (via Fusion, verified by pixels)
Fusion comps and tools	`fusion_comp`, `fusion_comp_inspect`, `fusion_tool`
Audio gain, fades, volume envelope, normalize	`audio_process` (offline on the WAV, then import or replace)
Render / export	`render_presets`, `render_options`, `render_add_job`, `render_jobs`, `render_quick_export`
Interchange (EDL, FCPXML, OTIO, DRT, AAF)	`timeline_export`, `timeline_import`
What can't the API do?	`ui_fallback_guide` gives click-by-click UI steps for the user

3. Rules that prevent damage

Destructive operations need confirm=true: deleting projects, timelines, clips or stills, overwriting, and resetting grades. Ask the user first and pass confirm only after they agree. The server makes a .drp backup before destructive batch operations.
Prefer dry_run=true first for batch edits and show the user the plan.
Workarounds (razor, trims, transitions, speed) create a new timeline. Tell the user its name. Never delete their original timeline to "clean up".
color_grade_drx and color_copy_grade replace the clip's grade. The previous grade is kept as a Claude backup <time> color version, so tell the user it exists.
Do not enable python_exec. It is off by design (allow_exec in config).

4. Color grading tips

Look first: frame_preview before and after.
color_grade_drx takes panel numbers.
- Brightness rides on gain.master. Gain r/g/b only change balance; uniform r=g=b does nothing.
- Saturation is 0–100, with 50 neutral.
- Place windows with center_x / center_y (frame fractions 0–1, y down) rather than pan/tilt: one pan unit ≈ 82 px, so pan 65 is already off-frame.
- A size-50 circle has a radius of about 288 px; soft1 softens the edge.
Only circle windows and primaries are render-verified. Qualifiers, HSL curves, linear, gradient and polygon windows are written but unverified. Check with frame_preview and tell the user if they did not render.
color_grade_lut is global, applies to one node and is display-referred. It suits looks (contrast, curves, hue-vs-sat, skin protection), but cannot mask a region.

5. Known limits (say so instead of trying)

These have no API on Resolve Free: tracking, adding nodes to an existing grade, OpenFX parameters, Fairlight mixer/automation, and Studio-only AI tools (Magic Mask, Smart Reframe, voice isolation…). Offer ui_fallback_guide steps or a workaround (Fusion masks, offline audio).

Version quirks are recorded in the repo's COVERAGE.md. Examples:

AppendToTimeline end frames are exclusive.
ExportLUT only works on the Color page.
New projects are not listed until saved.

6. How to report back to the user

After each task, reply in this shape (short, in the user's language):

Done: what changed, from diff_state or the actual values, not from what was requested.
Where: project, timeline (name the new timeline if a workaround made one) and the clips affected.
Check: the frame_preview image path for visual changes, and any ok: false or verified: null item with the reason.
Undo: how to go back. Examples: the original timeline is untouched, the old grade is in version Claude backup <time>, or the .drp backup path.

If something could not be done, say which limit from section 5 applies and offer the UI steps.

Setup and quirk details are in references/setup.md and references/quirks.md.

7. Troubleshooting

Symptom	Fix
`bridge not running` / `token mismatch`	Click Workspace → Scripts → ClaudeBridge again, then retry
`frame_preview` fails	The Resolve window is minimized. Ask the user to restore it
A grade or Fusion change does not show in renders	Re-apply through the typed tool (it commits Fusion comps); verify with `frame_preview`
A Studio-only tool is refused	Expected on Free. Explain, offer a UI or manual alternative

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

It worked
Partly
It didn't

What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free