Connect AI agents to your video library.
The Model Context Protocol lets your AI client search your Cutsio video library, read transcripts, and file footage into folders — no API key.
https://studio.cutsio.com/mcpIncluded with the Studio and Enterprise plans.
What you can ask it.
The agent turns a sentence into tool calls against your library.
“Find the drone shots from the Lisbon shoot.”
Searches titles, tags, AI summaries, notes, and transcript text.
“What's in the steam vent clip?”
Returns the summary, tags, duration, and status for that asset.
“When does she mention pricing?”
Returns transcript segments with timestamps.
“What campaigns do we have?”
Lists your collections.
“What's inside the Acme campaign?”
Lists that collection’s assets and the folder each one sits in.
“Make a folder for the approved finals.”
Creates it in your file storage, nested wherever you ask, and records the change.
“File these five clips under B-roll.”
Moves up to 25 assets into a folder and flags when an active review link covers it.
Connect in three steps.
Paste the endpoint, sign in to Cutsio, approve the consent screen. Every client follows the same path, and there is no key to copy.
- 1
Paste the endpoint
Add the /mcp URL as a custom connector or server entry.
- 2
Sign in to Cutsio
Use the account that belongs to the workspace you want to read.
- 3
Approve the consent screen
Grant assets:read for search, transcripts, and listings, plus assets:write for folder organizing. Both are revocable.
- Claude (web and Desktop) and ChatGPT add it as a custom connector. Claude Code, Cursor, and VS Code use the config below.
claude mcp add --transport http cutsio https://studio.cutsio.com/mcp{
"mcpServers": {
"cutsio": {
"url": "https://studio.cutsio.com/mcp"
}
}
}VS Code uses the same shape with a top-level servers key instead of mcpServers.
Six read tools. Two write tools.
The client discovers them automatically when it connects. Reads need the assets:read scope; folder organizing also needs assets:write.
search_assetsText search across titles, tags, AI summaries, notes, and transcripts. Returns id, title, status, duration, tags, summary, and a link into Cutsio.
get_assetOne asset by id, with the same fields and a link a person can open.
get_transcriptTimestamped segments in milliseconds, or the reason there is no transcript yet.
list_collectionsCollections (projects, campaigns, deliveries) with ids and titles.
list_collection_videosThe assets inside one collection, oldest first — the same fields as search_assets plus the folder each asset sits in.
list_foldersThe folder tree in your workspace’s file storage, with parent ids so the agent can navigate it.
create_folderCreates a folder, optionally nested inside another. Capped at 12 levels deep, with unique names among siblings.
move_video_to_folderFiles up to 25 assets into a folder, or out of every folder with a null destination. Warns when a review link covers the destination.
Writes stay narrow
- Two organizing tools: create a folder, file up to 25 videos per call into one
- No uploads, edits, clips, deletes, folder renames, or media downloads
- Every write is audit-logged and attributed to the person who approved the connection
- A separate assets:write scope and 15/min budget
- Only the workspace of the account that approved it
Not a frame reader
MCP search is text-based: titles, tags, AI summaries, notes, and transcripts. For frame-level visual search, use Visual Intelligence in the Cutsio app.
Questions.
Do I need an API key?
No. Paste the endpoint, sign in to Cutsio, and approve the consent screen. Authentication is OAuth with two scopes: assets:read for search and transcripts, assets:write for folder organizing.
Can the agent change anything?
It can organize, not edit. With the assets:write scope it can create folders and file videos into them — nothing else. No uploads, metadata edits, clips, deletions, media downloads, folder renames, or review-link changes. Every write is recorded in the workspace audit trail.
Does search look at the video frames?
No. MCP search is text-based — titles, tags, AI summaries, notes, and transcripts. Frame-level visual search lives in Visual Intelligence inside the Cutsio app.
Which clients work?
Claude on the web and Desktop, Claude Code, ChatGPT with connectors enabled by a workspace admin, Cursor, VS Code, OpenCode, Cline, Windsurf, Warp, Gemini, and any other client that supports remote MCP servers.
Can an agent see another workspace?
No. The connection only sees the workspace of the account that approved it, and that account must belong to exactly one workspace.
What are the limits?
60 read calls and 15 write calls per workspace per minute, counted per call. A search returns 10 results by default, 25 maximum. Connector search does not draw on AI credits yet — in-app search is metered against your balance, the connector is not, for now.
Read more: Visual Intelligence · Media Asset Management · MCP in video workflows
Connect your library to any agent.
Search, transcripts, and folder organizing — two revocable scopes, no API key, in about three minutes.