Skip to main content
Glama

Resolve MCP Server

Talk to your timeline.

AI-native remote editing and automation for DaVinci Resolve 21. Keep the editorial workflows—precise B-roll replacement, clip transforms, frame understanding and remote operation—and add native Resolve AI, inspectable state, safer edits, captions, transcripts, silence tightening, and multi-track podcast/interview workflows.

Current release: 2.2.0. The 2.0 editing workflows and 2.2 multi-track variant workflows have been live-tested on both Windows and macOS. Some less common Resolve-native AI and legacy operations remain unit-tested or partially exercised only. See validation and limitations.

Current inventory: 77 tools, 16 read-only MCP resources, and 5 workflow prompts.

What you can ask

  • “What's in this timeline?” — read project, tracks, clips and markers.

  • “Find the interview at 13:22.” — search at 802 elapsed seconds.

  • “Replace the second clip on V2 with TAKEOFF_03, video only.”

  • “What's in the current frame?” — Moondream caption or visual Q&A.

  • “Find shots containing an airplane.” — sample visible timeline frames with Moondream.

  • “Transcribe every interview in this bin.” — Resolve-native transcription.

  • “Caption this episode and give me an SRT.” — Resolve's speech recognition, read back as text.

  • “Cut the dead air out of this podcast.” — builds a tightened copy; the original is untouched.

  • “Remove the ums and the tangent about parking.” — text-based cuts into a new timeline.

  • “Keep all cameras and microphones in sync while tightening this interview.” — cuts every carried track together.

  • “Add chapter markers based on these transcript timestamps.” — supply chapter boundaries.

  • “Create a rough cut from these takes in this order.”

  • “Render this timeline for YouTube.” — local Quick Export, uploading disabled.

  • “Tell me what is currently rendering.” — read render jobs and progress.

The assistant still supplies editorial judgment. It cannot query Resolve's IntelliSearch index. Word-level, speaker-labelled clip transcripts need Resolve 21.1+; timeline captions work on supported Resolve 21 Studio builds.

Related MCP server: Resolve Claude MCP

Requirements

  • DaVinci Resolve Studio 21 on the same workstation, running with Preferences > System > General > External scripting using > Local.

  • A compatible 64-bit Python 3.10+ installation. Python and native Resolve scripting-library compatibility must be checked on your machine.

  • An MCP client supporting stdio or Streamable HTTP.

  • Optional Moondream API key, only for cloud vision tools.

  • Optional ffmpeg on PATH (or RESOLVE_MCP_FFMPEG), only for silence detection and tightening.

  • Required native AI packages installed through Resolve's Extras Download Manager.

Blackmagic's API documentation describes a Free/Studio superset, but some functions fail without Studio or required Extras. This project's external automation target is Studio; it does not promise full functionality on the free edition.

Install on macOS

git clone https://github.com/guycochran/resolve-mcp-server.git
cd resolve-mcp-server
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e .
resolve-mcp --doctor
resolve-mcp

Install on Windows (PowerShell)

git clone https://github.com/guycochran/resolve-mcp-server.git
cd resolve-mcp-server
py -3 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e .
.\.venv\Scripts\resolve-mcp.exe --doctor
.\.venv\Scripts\resolve-mcp.exe

Activation is optional. If PowerShell blocks launcher scripts, run the executable directly.

Supported launchers: resolve-mcp, resolve-mcp-server, python -m resolve_mcp, python src/server.py, start.sh, and start.ps1. The existing FastMCP implementation is retained using the maintained MCP SDK 1.x line (mcp>=1.28,<2); SDK 2.x is a separate migration.

Automatic scripting discovery

Platform

Default Modules directory

Windows

%PROGRAMDATA%\Blackmagic Design\DaVinci Resolve\Support\Developer\Scripting\Modules

macOS

/Library/Application Support/Blackmagic Design/DaVinci Resolve/Developer/Scripting/Modules

Override with PYTHONPATH_RESOLVE (Modules directory) or RESOLVE_SCRIPT_API (parent Scripting directory). Blackmagic's loader honors RESOLVE_SCRIPT_LIB for a nonstandard native library location. Linux's standard path is retained as a best effort, without claiming Linux testing.

Connection health checks are cached for five seconds; failed attempts have a two-second cooldown. resolve_reconnect forces an immediate retry. --doctor separates OS process presence from scripting connectivity and reports the version/edition when the API is available.

MCP client configuration

For a client using an mcpServers JSON configuration, use an absolute interpreter path and module arguments. Windows example:

{
  "mcpServers": {
    "resolve": {
      "command": "C:/path/to/resolve-mcp-server/.venv/Scripts/python.exe",
      "args": ["-m", "resolve_mcp"],
      "env": {"TRANSPORT": "stdio"}
    }
  }
}

On macOS use /absolute/path/to/resolve-mcp-server/.venv/bin/python. No working-directory assumption is needed after installation.

Copy .env.example to .env in a source checkout for local configuration. Environment variables take precedence. Installed-wheel deployments can set RESOLVE_MCP_ENV_FILE to an explicit private file path. Do not commit credentials.

Remote operation: authenticated Streamable HTTP

Remote MCP client → HTTPS access gateway → localhost:3001/mcp → Resolve
Local MCP client  → stdio                                  → Resolve

Set TRANSPORT=http, MCP_AUTH_TOKEN and optionally PORT. The server defaults to HOST=127.0.0.1. Generate a random token, for example with python -c "import secrets; print(secrets.token_urlsafe(32))", and save it in your private environment configuration. A token requires at least 32 characters.

Clients send Authorization: Bearer <token> on every HTTP request. External binding requires a token; setting MCP_PUBLIC_URL also requires one. Use MCP_PUBLIC_URL=https://resolve.example.com to allow your exact gateway host and origin. Invalid host/origin requests remain blocked.

This is shared-token authentication for trusted operators, not an OAuth login server. Clients requiring OAuth need a compatible authentication gateway. A token grants access to all exposed tools and local media operations. Run one server process per Resolve instance, with no concurrent GUI edits during mutations. TLS and operator access policies belong at the gateway.

Cloudflare Tunnel

Route a named tunnel hostname to http://127.0.0.1:3001. Set MCP_PUBLIC_URL to that HTTPS hostname, keep the backend on loopback, and retain bearer authentication. Apply Cloudflare Access policies appropriate for the operators/clients; an Access login page alone is not compatible with every MCP client. Configure service credentials or an OAuth-capable gateway where necessary.

See Cloudflare's published application documentation. A tunnel supplies connectivity; configure access controls as a separate step.

Tailscale

Keep the same loopback backend and bearer token. Use Tailscale Serve to proxy http://127.0.0.1:3001 over your tailnet's HTTPS hostname; set MCP_PUBLIC_URL to that origin and restrict operators with tailnet policy. See Tailscale Serve.

Do not forward the workstation's HTTP port directly onto the public internet. No arbitrary Python or Lua execution tool is exposed.

Two complementary kinds of AI

Resolve-native AI

Transcription with optional speaker detection, transcription clearing, audio classification/clearing, IntelliSearch analysis/reset, Slate ID markers, motion deblur, speech generation, and session-wide background-task disabling.

Native AI runs through Resolve. IntelliSearch and Slate ID need their Extras packages; speech generation needs AI Speech Generator. Face identification is off by default. Folder transcription/classification includes nested folders. Resetting IntelliSearch affects the whole project. Background-task disabling lasts for the Resolve session and has no API enable counterpart.

Moondream visual-language analysis

Set MOONDREAM_API_KEY to preserve frame descriptions, object detection and visual Q&A. These requests send compressed frames to Moondream's cloud API. Images use unique temporary paths and are deleted after each request; JPEG compression happens in memory. Review Moondream API documentation.

Visual shot search samples one midpoint frame per timeline clip, up to the requested limit, and restores page/playhead. It observes the visible composite, including upper tracks; it can miss objects outside the sample. It is not an exhaustive source-media search or a query into native IntelliSearch.

Read-only MCP resources

Each resource returns JSON: {success: true, data: ...} or {success: false, error: {code, message}}. Reads never select a project/bin/timeline.

Area

Resource URIs

System

resolve://system/status

Project

resolve://project/current, resolve://project/list, resolve://project/settings, resolve://project/timelines

Timeline

resolve://timeline/current, resolve://timeline/tracks, resolve://timeline/items, resolve://timeline/markers

Media

resolve://mediapool/folders, resolve://mediapool/current-folder, resolve://mediapool/clips

Render

resolve://render/jobs, resolve://render/formats, resolve://render/presets, resolve://render/is-rendering

Project listing is scoped to the current database folder; media clip listing is scoped to the current bin. Folder paths and workflow searches can traverse all bins.

Editorial workflows and recovery

All 53 original tool names remain. Newer workflows include media/timeline lookup, resolve_insert_broll, resolve_build_rough_cut, marker/chapter helpers, resolve_render_for_youtube, visual shot search, captions/transcripts, silence analysis, reviewable cut variants, and render waiting.

Clip replacement

resolve_replace_clip still reads record position, deletes without ripple, and inserts at that same record frame. Video-only is the default.

  • Use 1-based track/clip indices. dry_run=true returns the plan without mutations.

  • Media names must be unique; new_media_id disambiguates them.

  • Append source ranges are half-open: [in, out_exclusive); automatic matching uses in + duration. The existing source_end_frame argument is exclusive (zero means automatic). Plans return source_out_exclusive; original source_in_native / source_out_native preserve raw TimelineItem readbacks and must not be reused as append bounds.

  • Source bounds, timeline position and track locks are checked before deletion.

  • Mixed source/timeline FPS is rejected rather than guessed.

  • A full recovery timeline is created before changing clips.

  • Linked peers are unlinked before deleting only the target, then linked to the replacement. Interview audio is not included in the delete call.

  • An insertion, duration or relinking failure selects the recovery timeline. The modified original remains for inspection; references to that timeline are not automatically redirected. A recovery copy is not an atomic undo.

  • The new clip uses its own media defaults. Original effects, grades, Fusion and retiming remain in the recovery copy.

media_type=2 explicitly targets an audio-track item. Combined video/audio replacement is rejected because a single index cannot safely identify both targets. This is a safety-related change from the permissive legacy parameter.

B-roll insertion requires an empty destination interval. B-roll and rough-cut tools default to dry-run. Chapter markers take supplied timestamps; they do not infer speaker changes. Direct trim/move operations are deferred because the API does not provide a general, reliable in-place edit with full effect preservation.

Transcripts, captions and multi-track tightening

These workflows are designed to preserve the source timeline by reading it or building a new variant.

Tool

What it does

resolve_create_captions

Runs Resolve's auto-captioning on the current timeline and verifies success by reading cues back.

resolve_get_transcript

Returns caption cues, or on Resolve 21.1+ a clip's native transcript with speakers/word times. Can write SRT, VTT or text to a new path.

resolve_detect_silence

Read-only ffmpeg silencedetect, calibrated from recording noise floor unless overridden. Can consider all carried audio tracks.

resolve_tighten_silence

Removes shared dead air into a new timeline, keeping a small pause around speech. Dry-run by default.

resolve_build_cut_variant

Removes specified time ranges across carried tracks into a new timeline. Dry-run by default.

resolve_wait_for_render

Waits for a render job or queue, then reports status and output file.

In 2.2, variant editing is multi-track aware. The source timeline is duplicated, the copy is emptied and rebuilt from kept media ranges on their original track numbers. Cameras and separate microphone tracks are cut together, links are restored, timeline markers in kept time move with the edit, and each track is read back and verified against the plan.

Silence is cut only where the selected audio set is quiet. With the default carried-track mode, a guest answering while the host is silent is not mistaken for dead air. Track names, audio formats, enable states and timeline settings are carried where supported.

Locked, non-empty tracks are intentionally refused rather than automatically unlocked because Resolve lock behavior differs across tested 21.x builds and can affect duplicate/source timelines. Titles, generators, compound/multicam clips, retimed clips and media at a different frame rate cannot always be rebuilt exactly; unsupported cases are rejected or reported instead of guessed.

Free-edition calls to Resolve's AI features can open a modal upgrade dialog that disrupts later API calls, so these tools refuse to run unless the product is Resolve Studio.

Live validation highlights

Windows

Resolve Studio 21.0.4.5 / Python 3.12.14. The 2.2 multi-track suite passed 34/34 checks, including locked-track refusal, shared-silence detection across A1/A2/A3, synchronized V1/A1/A2/A3 tightening, link restoration, marker movement, text cuts, source-timeline preservation, and render/wait.

macOS

Apple silicon / macOS 27.0 / Resolve Studio 21.1.0.17 / Python 3.14.2. The 2.2 live run passed, including shared-silence detection, multi-track tightening, link restoration, marker mapping, text cuts, render/wait and project save. The macOS unit suite reported 164 tests.

See docs/VALIDATION.md for fixtures, raw values and Resolve-version-specific findings.

Workflow prompts

MCP clients that support prompts get ready-made recipes with the safety rules built in: podcast_episode_edit, tighten_recording, captions_and_transcript, safe_shot_replacement and youtube_delivery.

Development

Codex and other coding agents should read AGENTS.md before making changes.

python -m pip install -e ".[dev]"
python -m pytest
python -m ruff check src tests
python -m build

See architecture, manual integration tests, validation, the 2.0 implementation report, the 2.1 release notes and 2.2 release notes.

License

MIT. Resolve-native additions were implemented independently from Blackmagic's installed scripting documentation. No source was copied from the Digital Workflow Company reference implementation. The 2.1 transcript and tightening workflows were written independently; the idea of caption readback and calibrated silence detection was informed by reviewing samuelgursky/davinci-resolve-mcp (MIT).

Available Tools

77 tools
resolve_add_fusion_compA

Add a new Fusion composition to the clip at the playhead.

Inputs: no named parameters. Returns: result

resolve_add_markerB

Add timeline marker. Legacy frame=0 selects playhead; other values are timeline-relative. Use create_chapter_markers to explicitly address frame zero.

Inputs: name, note, color, frame, +1 more. Returns: result

resolve_add_marker_at_playheadA

Add a marker at the exact current playhead, including drop-frame timecode support.

Inputs: name (required), note, color.

resolve_add_render_jobB

Add a custom render job to the render queue.

Args: output_dir: Output directory path. filename: Custom filename (optional). format: Video format (e.g., "mov", "mp4"). Leave empty for project default. codec: Codec name (e.g., "ProRes422", "H264"). Leave empty for project default. width: Output width. 0 = use project setting. height: Output height. 0 = use project setting. frame_rate: Output FPS. 0 = use project setting.

Inputs: output_dir (required), codec, width, format, +3 more. Returns: result

resolve_analyze_intellisearchA

Analyze a clip or folder for IntelliSearch. Needs Faster/Better Extras. Face identification is opt-in. This starts native analysis; it does not query a search index.

Inputs: clip_name, better_mode, folder_path, identify_faces.

resolve_analyze_slateA

Run native Slate ID analysis on a clip or folder and add slate markers. Requires AI Slate ID Extras.

Inputs: clip_name, folder_path, marker_color.

resolve_append_to_timelineA

Add a clip from the media pool to the current timeline.

Args: clip_name: Name of the clip in the media pool. track_index: Target video track (1-based, default: 1). media_type: 0 = video+audio (default), 1 = video only, 2 = audio only. Use 1 for b-roll to preserve interview audio.

Inputs: clip_name (required), media_type, track_index. Returns: result

resolve_apply_lutA

Apply a LUT to a clip's node graph. Targets the clip at the playhead.

Must be on the Color page. Common built-in LUTs:

  • "Film Looks/Rec709 Kodak 2383 D65.cube"

  • "Film Looks/Rec709 Fujifilm 3513DI D65.cube"

Args: lut_path: LUT file path (relative to Resolve's LUT directory or absolute). node_index: Node to apply LUT to (1-based, default: 1).

Inputs: lut_path (required), node_index. Returns: result

resolve_ask_about_frameB

Send the current frame to Moondream for visual question answering.

Inputs: question (required). Returns: result

resolve_build_cut_variantA

Text-based / manual editing: cut the listed ranges from ALL carried tracks into a NEW timeline (dry-run by default). remove: [{start_seconds, end_seconds}] in timeline-relative seconds, e.g. caption cues from resolve_get_transcript(source="captions") for filler words, false starts or off-topic passages. Ranges may overlap; the original timeline is not changed. tracks="all" (default) carries every video/audio track, keeping mics and cameras in sync; tracks="spine" carries only the spine track (A1 by default).

Inputs: remove (required), tracks, dry_run, open_variant, +5 more.

resolve_build_rough_cutA

Assemble complete media clips in supplied order into a NEW timeline. Rejects ambiguous media and existing timeline names. Defaults to planning only. This is an ordered assembly; it does not infer cuts from a transcript.

Inputs: name (required), clip_names (required), dry_run.

resolve_classify_audioB

Classify audio for one clip, or the current/specified folder AND nested folders.

Inputs: clip_name, folder_path.

resolve_clear_audio_classificationC

Remove audio classification for one clip, or the current/specified folder AND nested folders.

Inputs: clip_name, folder_path.

resolve_clear_transcriptionB

Delete transcription for one clip, or the current/specified folder AND nested folders.

Inputs: clip_name, folder_path.

resolve_create_binC

Create a new bin (subfolder) in the media pool.

Args: name: Name for the new bin.

Inputs: name (required). Returns: result

resolve_create_captionsA

Auto-caption the CURRENT timeline with Resolve's speech recognition (Studio, adds a subtitle track). Clips and audio are not changed. The native true/false result is unreliable, so success means a new subtitle track or new cues were read back. Then use resolve_get_transcript(source="captions"). language: auto, english, spanish, ...; preset: default, netflix, teletext; chars_per_line 0 = preset default.

Inputs: preset, dry_run, language, gap_frames, +2 more.

resolve_create_chapter_markersA

Create markers from supplied chapter/speaker boundaries. Each entry: {frame: timeline-relative integer, name: string, note: optional string}. Validates the complete batch before writing. Does not extract transcripts or infer speaker times.

Inputs: chapters (required), color.

resolve_create_color_versionB

Create a new color version on the clip at the playhead.

Args: name: Name for the color version. version_type: 0 = local (default), 1 = remote.

Inputs: name (required), version_type. Returns: result

resolve_create_compound_clipB

Create a compound clip from a 1-based video-track range. Creates a recovery timeline first.

Inputs: name, end_clip, start_clip, track_index. Returns: result

resolve_create_projectB

Create a new project.

Args: name: Name for the new project.

Inputs: name (required). Returns: result

resolve_create_timelineB

Create a new empty timeline.

Args: name: Name for the new timeline.

Inputs: name (required). Returns: result

resolve_delete_clipA

Delete exactly one clip; defaults to non-ripple. Linked peers remain. Creates a recovery copy.

Inputs: ripple, clip_index, track_type, track_index. Returns: result

resolve_delete_markersA

Delete markers by explicit color, All, or exact timeline-relative frame (including zero). No arguments make no changes. Returns the removed marker records for restoration.

Inputs: color, frame. Returns: result

resolve_describe_frameA

Send the current frame to Moondream for a caption. detail: short or normal.

Inputs: detail. Returns: result

resolve_detect_in_frameA

Send the current frame to Moondream to detect described objects and return bounding boxes.

Inputs: object_description (required). Returns: result

resolve_detect_silenceA

Find dead air on the current timeline with ffmpeg (read-only; needs ffmpeg). detect_on="carried" reports only time when EVERY enabled audio track is quiet (safe for multi-mic podcasts); "spine" analyzes just the spine track. threshold_db 0 = calibrate per clip. Returns timeline-relative seconds (the same scale as resolve_get_transcript captions).

Inputs: tracks, detect_on, audio_stream, threshold_db, +3 more.

resolve_disable_background_tasksA

Disable ALL Resolve background tasks for this application session. The API has no corresponding enable call; restart Resolve to reset this setting.

Inputs: no named parameters.

resolve_duplicate_timelineC

Duplicate the current timeline.

Args: new_name: Name for the duplicate (optional, defaults to original + " Copy").

Inputs: new_name. Returns: result

resolve_export_lutA

Export a LUT from the current clip's grade.

Args: output_path: Absolute path for the output .cube file. export_type: "17pt", "33pt" (default), or "65pt".

Inputs: output_path (required), export_type. Returns: result

resolve_export_timelineC

Export the current timeline to an interchange format.

Args: output_path: Absolute path for the output file. format: Export format — "fcpxml" (default), "edl", "csv", "aaf", "otio".

Inputs: output_path (required), format. Returns: result

resolve_find_media_clipA

Find shots by case-insensitive name and/or metadata substrings across all bins. Returns exact names, media IDs and folder paths for unambiguous editing. Not visual search.

Inputs: limit, query, metadata.

resolve_find_shots_by_visual_descriptionA

Look for objects such as an airplane in timeline shots using Moondream. Sends ONE midpoint timeline frame per clip (up to 50) to the cloud; may miss objects elsewhere. Analyzes the visible composite, including overlays/upper tracks. Temporarily moves playhead/page. Reports sampled clip indices, matches and restoration status; not exhaustive source-media search.

Inputs: object_description (required), max_clips, track_index.

resolve_find_timeline_clipA

Find clips by name and/or position. timecode is absolute HH:MM:SS:FF; seconds is elapsed since timeline start (13:22 = 802 seconds). End frame is exclusive. track_index 0 searches all tracks of the selected type.

Inputs: query, seconds, timecode, track_type, +1 more.

resolve_generate_speechA

Generate a speech clip into the media pool, leaving the timeline untouched. Requires AI Speech Generator Extras. Text is limited to 350 characters.

Inputs: text (required), filename (required), voice.

resolve_get_clip_propertiesA

Get detailed properties of a media pool clip.

Args: clip_name: Name of the clip in the media pool.

Returns JSON with all clip properties (duration, resolution, codec, fps, etc.).

Inputs: clip_name (required). Returns: result

resolve_get_clip_transformA

Read clip transform values at the playhead or explicit 1-based video track/clip indices.

Inputs: clip_index, track_index. Returns: result

resolve_get_current_pageA

Get the currently active DaVinci Resolve page.

Returns the page name: media, cut, edit, fusion, color, fairlight, or deliver.

Inputs: no named parameters. Returns: result

resolve_get_current_timelineA

Get detailed info about the current timeline.

Returns JSON with name, timecode, track counts, and start/end frames.

Inputs: no named parameters. Returns: result

resolve_get_fusion_compsA

List Fusion compositions on a timeline clip. Targets clip at playhead by default.

Args: track_index: Video track (1-based). 0 = use playhead. clip_index: Clip position on track (1-based). 0 = use playhead.

Returns JSON with comp names and count.

Inputs: clip_index, track_index. Returns: result

resolve_get_fusion_toolsA

List all tools in a Fusion composition on the clip at the playhead.

Args: comp_index: Fusion comp index (1-based, default: 1).

Returns JSON array of tools with their type and name.

Inputs: comp_index. Returns: result

resolve_get_lutA

Get the LUT currently applied to a node on the clip at the playhead.

Args: node_index: Node index (1-based, default: 1).

Inputs: node_index. Returns: result

resolve_get_markersA

Read timeline markers keyed by timeline-relative frame.

Inputs: no named parameters. Returns: result

resolve_get_playheadA

Get the current playhead timecode position.

Inputs: no named parameters. Returns: result

resolve_get_project_settingsA

Get all settings for the current project.

Returns JSON with frame rate, resolution, color science, and other project settings.

Inputs: no named parameters. Returns: result

resolve_get_render_statusA

Get the status of all render jobs.

Returns JSON with each job's status, progress percentage, and completion state.

Inputs: no named parameters. Returns: result

resolve_get_statusA

Get current DaVinci Resolve status including version, current project, timeline, and page.

Returns JSON with product name, version, current page, project name, timeline name, and connection state.

Inputs: no named parameters. Returns: result

resolve_get_track_itemsB

List all clips on a specific track.

Args: track_type: "video", "audio", or "subtitle" (default: "video"). track_index: Track number, 1-based (default: 1).

Returns JSON array of clips with name, start/end frame, duration, and source info.

Inputs: track_type, track_index. Returns: result

resolve_get_transcriptA

Read transcript TEXT with timing. source="captions": cues from the current timeline's subtitle track(s) (0 = all); seconds are timeline-relative, frames absolute. Works on any Resolve 21 Studio after resolve_create_captions. source="clip": a media pool clip's full native transcript (Resolve 21.1+) with speakers and word times, in seconds from the clip's start; run resolve_transcribe_audio first. format json|srt|vtt|text. output_path (absolute, new file) writes SRT/VTT/text to disk.

Inputs: format, source, media_id, clip_name, +3 more.

resolve_import_mediaA

Import media files into the current media pool bin.

Args: file_paths: List of absolute file paths to import (e.g., ["/Users/guy/video.mov"]).

Returns names of successfully imported clips.

Inputs: file_paths (required). Returns: result

resolve_insert_brollA

Insert video-only B-roll into an EMPTY interval on an existing video track. Absolute record frame; defaults to dry-run. Rejects overlap, locked tracks and FPS mismatch. Creates a recovery copy before an actual insertion.

Inputs: clip_name (required), record_frame (required), duration_frames (required), dry_run, +3 more.

resolve_insert_titleA

Insert Text+ at playhead. Validates size/position and reports partial failures after creation.

Inputs: text (required), font_size, position_x, position_y. Returns: result

resolve_list_color_versionsA

List all color versions on the clip at the playhead.

Args: version_type: 0 = local versions (default), 1 = remote versions.

Returns JSON with version names and current version.

Inputs: version_type. Returns: result

resolve_list_mediaA

List clips in the media pool. Shows the current bin by default.

Args: folder_path: Optional subfolder name to navigate to (e.g., "B-Roll"). Leave empty for current folder.

Returns JSON with folder name, clips (with metadata), and subfolders.

Inputs: folder_path. Returns: result

resolve_list_projectsA

List all projects in the current database folder.

Returns a JSON array of project names.

Inputs: no named parameters. Returns: result

resolve_list_render_presetsA

List all available render presets (both standard and Quick Export).

Returns JSON with standard render presets and Quick Export presets.

Inputs: no named parameters. Returns: result

resolve_list_timelinesA

List all timelines in the current project.

Returns JSON array with each timeline's name, index, duration, and track counts.

Inputs: no named parameters. Returns: result

resolve_load_color_versionA

Switch to a color version by name on the clip at the playhead.

Args: name: Color version name. version_type: 0 = local (default), 1 = remote.

Inputs: name (required), version_type. Returns: result

resolve_load_projectC

Open a project by name.

Args: name: Project name to load.

Inputs: name (required). Returns: result

resolve_modify_title_textB

Modify Text+ on the playhead clip or explicit 1-based video track/clip indices.

Inputs: new_text (required), clip_index, track_index. Returns: result

resolve_open_pageA

Switch DaVinci Resolve to a specific page.

Args: page: Page name — "media", "cut", "edit", "fusion", "color", "fairlight", or "deliver"

Inputs: page (required). Returns: result

resolve_quick_exportB

Render the current timeline using a Quick Export preset.

Common presets: "H.264 Master", "H.265 Master", "ProRes 422 HQ", "YouTube", "Vimeo", "TikTok", "Twitter".

Args: preset: Quick Export preset name. output_dir: Output directory (optional, uses project default if empty). filename: Custom filename (optional, uses timeline name if empty).

Inputs: preset (required), filename, output_dir. Returns: result

resolve_reconnectA

Force reconnection to DaVinci Resolve. Use if Resolve was restarted.

Inputs: no named parameters. Returns: result

resolve_remove_motion_blurA

Create new deblurred media from ONE uniquely named clip; does not replace timeline clips. filename names the native output. Format/codec availability depends on this workstation.

Inputs: filename (required), clip_name (required), codec, format, +2 more.

resolve_render_for_youtubeA

Render locally with an installed Quick Export preset; explicitly disables uploading to YouTube. Requires an existing absolute output directory and a filename. Returns Resolve's actual status.

Inputs: filename (required), output_dir (required), preset.

resolve_replace_clipA

Replace at the original record position without ripple. Video-only preserves linked audio. Indices are 1-based. media_type 1 targets video; 2 targets audio. Combined replacement is rejected. source_end_frame is EXCLUSIVE; zero auto-matches source_start_frame + original duration. Source FPS must match timeline. Native source-out readbacks are reported separately. Validates bounds/locks/ambiguity and creates a recovery timeline before mutation. dry_run returns the plan without edits; new_media_id disambiguates duplicate names.

Inputs: clip_index (required), track_index (required), new_clip_name (required), dry_run, +4 more. Returns: result

resolve_reset_intellisearchA

Clear IntelliSearch analysis for the ENTIRE current project (Resolve API scope).

Inputs: no named parameters.

resolve_save_projectB

Save the current project.

Inputs: no named parameters. Returns: result

resolve_set_clip_enabledA

Enable/disable a clip at playhead or explicit 1-based video track/clip indices.

Inputs: enabled (required), clip_index, track_index. Returns: result

resolve_set_clip_speedC

Legacy speed-property request. Resolve may reject it; timeline retiming is not guaranteed by the API. This acts on the underlying media-pool property, potentially affecting other uses of the media.

Inputs: speed (required), clip_index, track_index. Returns: result

resolve_set_clip_transformA

Set pan/tilt, zoom (0–100), rotation (-360–360), opacity (0–100). Targets playhead, or explicit 1-based video track/clip indices. Validates before writing. Returns JSON including original values and any partial failure.

Inputs: pan, tilt, zoom_x, zoom_y, +4 more. Returns: result

resolve_set_current_timelineA

Switch to a different timeline by name or index.

Args: name: Timeline name to switch to (preferred). index: Timeline index (1-based). Used if name is empty.

Inputs: name, index. Returns: result

resolve_set_playheadA

Set the playhead to a specific timecode.

Args: timecode: Timecode string (e.g., "01:00:05:00").

Inputs: timecode (required). Returns: result

resolve_set_project_settingC

Set a project setting.

Args: setting: Setting name (e.g., "timelineFrameRate", "timelineResolutionWidth"). value: Value to set.

Inputs: value (required), setting (required). Returns: result

resolve_start_renderA

Start rendering all queued render jobs.

Inputs: no named parameters. Returns: result

resolve_tighten_silenceA

Remove dead air from a talking-head/podcast timeline into a NEW timeline (dry-run by default). A pause is cut only where every enabled mic track is quiet for longer than min_silence_seconds; keep_pause_seconds of room stays at each side. All tracks are cut together, so cameras and mics stay in sync; timeline markers in kept time move with the edit. Needs ffmpeg. tracks="all" (default) carries every video/audio track, keeping mics and cameras in sync; tracks="spine" carries only the spine track (A1 by default).

Inputs: tracks, dry_run, detect_on, audio_stream, +9 more.

resolve_transcribe_audioA

Transcribe one exact clip, or a folder AND nested folders when clip_name is empty. folder_path is an exact resource path; empty uses current bin. Requires Resolve 21. Returns operation status, not a transcript; speaker detection is optional.

Inputs: clip_name, folder_path, use_speaker_detection.

resolve_wait_for_renderA

Wait until a render job (or all rendering, if job_id is empty) finishes, then report the final status and output file. Other MCP calls wait while this runs, so keep timeouts modest (max 3600 s). stop_on_timeout=True calls StopRendering if the deadline passes.

Inputs: job_id, poll_seconds, stop_on_timeout, timeout_seconds.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.1/5.0

    Scored across 77 tools

    Disambiguation3/5

    Several tools have close cousins: add_marker overlaps with add_marker_at_playhead, quick_export overlaps with render_for_youtube, and there are multiple transcription/analyze commands. The detailed descriptions usually clarify the intended target, but with 77 tools an agent has real misselection risk.

    Naming Consistency4/5

    The resolve_ prefix and verb_noun pattern are used very consistently across the entire set. Minor deviations like resolve_reconnect, resolve_render_for_youtube, and resolve_wait_for_render break the strict verb_noun pattern, but they are still predictable.

    Tool Count2/5

    77 tools is well beyond the 25+ over-scoped threshold. The set bundles media management, editing, color, Fusion, render, and AI vision/audio features into one server, making the namespace heavy to navigate even though each tool has a defined job.

    Completeness3/5

    The surface is impressively broad for editing, color, audio, and rendering, but lifecycle gaps remain: no delete/rename project or timeline, and no trim/transition/move operations for timeline clips. Agents can complete many workflows but may hit dead ends on routine cleanup/edit tasks.

    Related MCP Connectors

    Related MCP Servers