Model Context Protocol•

Connect your Agent

Give Claude Desktop, OpenAI, Cursor, Claude Code and autonomous AI agents native access to video cuts, reviewer comments, spatial canvas drawings and DaVinci Resolve / Premiere Pro timeline markers.

Step 1: Choose your agent client

In Claude Desktop Settings → Connectors → Add MCP Server, enter the server URL. Claude initiates an in-browser authorization approval for zero-terminal setup:

https://dusken.io/api/mcp

Or add to claude_desktop_config.json:

{
  "mcpServers": {
    "dusken": {
      "url": "https://dusken.io/api/mcp"
    }
  }
}

Add to ~/.cursor/mcp.json or Cursor Settings → Features → MCP:

{
  "mcpServers": {
    "dusken": {
      "url": "https://dusken.io/api/mcp",
      "headers": {
        "Authorization": "Bearer <WORKSPACE_API_KEY>"
      }
    }
  }
}

Works out of the box with any MCP-compatible environment, including Windsurf (mcp_config.json), Zed (settings.json), Cline and Continue.dev:

MCP Server Configuration:

{
  "mcpServers": {
    "dusken": {
      "url": "https://dusken.io/api/mcp",
      "headers": {
        "Authorization": "Bearer <WORKSPACE_API_KEY>"
      }
    }
  }
}

Send standard JSON-RPC 2.0 requests over HTTP POST:

curl -X POST https://dusken.io/api/mcp \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <WORKSPACE_API_KEY>" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Step 2: Verify the connection

Paste this prompt into your agent chat to verify tools and workspace permissions:

“Connect to Dusken MCP, list the cuts in my workspace and summarize all reviewer feedback on the latest cut.”

Capabilities

What your agent can do in Dusken

Read Review Notes & Transcripts

Access frame-anchored reviewer comments, threaded replies, status flags and spoken dialogue audio transcripts with millisecond timecodes.

Inspect Spatial Drawing Arrows

Extract on-canvas arrows and freehand shapes translated into human spatial quadrants (top-left, lower-third) with high-res composite frames.

Export NLE Timeline Markers

Generate DaVinci Resolve EDL, Adobe Premiere Pro CSV/XML, or Final Cut Pro marker files so editors can drop reviewer requests directly onto their timeline.

Autonomous Render Loops

Use wait_for_feedback long-polling to sleep until human notes arrive, then auto-render, upload and stack the next version.

Reference

Prompts & Tool Reference

Complete specification of pre-engineered prompts and callable endpoints exposed by the Dusken MCP server.

Pre-Engineered Workflow Prompts

Callable via standard MCP prompts/get with direct argument injection.

summarize_cut_feedbackArgs: asset_id (required), version_id (optional)

Fetches all timecoded comments, reviewer voice notes and frame drawing annotations for a cut. Synthesizes feedback into four distinct review buckets: Pacing & Editorial, Visual & Color, Audio & Music and an Action Items Checklist with exact timecodes and deep links.

compare_cuts_diffArgs: asset_id (req), v1_id (req), v2_id (req)

Performs an automated diff between Cut V1 and Cut V2. Highlights resolved client notes versus new revision requests, enabling editors to verify that client notes were completely addressed before submitting the next cut.

client_notes_to_timelineArgs: asset_id (required), notes_text (required)

Ingests unstructured feedback from external client emails or Slack messages, extracts timecode mentions and converts them into structured, timestamped review comments directly inside Dusken.

export_to_nleArgs: asset_id (required), format: "resolve" | "premiere" | "fcpx"

Exports client feedback formatted as native timeline markers for DaVinci Resolve (EDL), Adobe Premiere Pro (XML/CSV) and Final Cut Pro X, allowing video editors to view reviewer notes directly on their edit track.

Callable MCP Tools (14 Endpoints)

Invoked by connected agents via standard tools/call JSON-RPC requests.

Content Discovery & Navigation

list_workspaces

Lists all workspaces accessible by the current authentication token with workspace IDs, names and team roles.

list_folders

Lists folders and project bins within a workspace, allowing agents to navigate deep folder hierarchies.

get_contents

Returns all video cuts, media assets and child folders inside a specified folder, including version counts and metadata.

get_asset

Fetches comprehensive metadata for a specific video cut, including resolution, duration, frame rate (FPS) and complete version stack.

Feedback & Canvas Annotations

list_feedback

Returns all timecoded comments, replies, voice note transcripts and resolution states for a video cut or version.

get_feedback

Fetches a single feedback thread by ID, including nested reply threads, user avatars and attached timestamps.

get_annotated_frames

Extracts on-canvas vector drawings made during review. Automatically describes quadrant positions (top-left, lower-third), delta arrow directions and attaches high-resolution composite JPEG frames.

search_transcription

Performs full-text search across automated speech-to-text transcripts, returning matching phrases with exact millisecond timestamps.

post_feedback

Allows agents to submit new comments or threaded replies anchored to a specific video timestamp with status tags.

update_feedback

Updates comment text or marks reviewer action items as completed once an editorial revision has been applied.

wait_for_feedback

Long-polling webhook loop (up to 25s) that holds the HTTP connection until human reviewers submit new comments, enabling autonomous render and revision loops.

Media Ingestion & Video Uploads

get_asset_status

Inspects Mux encoding status, HLS stream readiness and automated audio transcription progress for uploaded media.

upload_video

Ingests a new video asset from a public URL or render pipeline output into a target workspace folder.

upload_file_start & upload_file_complete

Multipart S3/Cloudflare R2 direct upload tools designed for large master video file transfers.

NLE Markers & Version Diffs

export_nle_markers

Generates marker export files formatted for DaVinci Resolve (EDL) and Adobe Premiere Pro (XML/CSV) containing all reviewer notes.

compare_versions

Diffs reviewer comments across Cut V1 and Cut V2, tracking which review notes were resolved and identifying newly introduced items.