mcp.film / Editing & Post / TwelveLabs Jockey
TwelveLabs Jockey
Cross-video agent MCP: build a knowledge store, then ask your whole library questions.
What it does
Jockey is TwelveLabs' agentic layer over a corpus rather than a clip. You create a knowledge store, add video and images to it, and then ask questions that span the whole collection — Jockey decomposes the query, retrieves, segments and reasons across items, and answers with clip citations rather than a wall of prose. For an editor this is the archive researcher: find every shot of a location across three seasons of footage, track a face or object through unrelated rushes, or have it assemble a highlight lane before anyone opens the NLE. It is a hosted OAuth endpoint, so there is nothing to install and no key to manage.
Connect
https://mcp.twelvelabs.io/jockey/mcp
claude mcp add --transport http --scope user jockey https://mcp.twelvelabs.io/jockey/mcp
{
"mcpServers": {
"twelvelabs-jockey": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.twelvelabs.io/jockey/mcp"
]
}
}
}
{
"mcpServers": {
"twelvelabs-jockey": {
"url": "https://mcp.twelvelabs.io/jockey/mcp"
}
}
}
Field notes
Still a research preview: the Jockey product page says it is "currently accepting limited sign-ups", so a free account is not a guaranteed seat. Pricing is now published there, metered by knowledge-store size: Free at 5 GB, Plus at $20/month for 100 GB, Pro at $100/month for 500 GB, with rate limits rising by tier. A real season of rushes outgrows the free tier quickly. Tool names are null because TwelveLabs does not publish a catalogue — the docs tell you to ask the connected client to list them, so run `tools/list` and see what you get rather than trusting any third-party listing. Hard limits are 4 GB per video and 32 MB per image, video URLs must be publicly reachable, and ingestion is asynchronous: adding a corpus is a wait, not a call. OAuth sessions can expire, and the docs tell you to re-authenticate through `/mcp` when that happens. Clients without widget support get a text fallback instead of the gallery/timeline/highlight-reel UI, which in practice means the citations are still usable but the assembly view is not. Requires a client that supports remote HTTP servers with OAuth. The docs name Claude and Claude Code, and the product page says ChatGPT support is "coming soon". Re-checked 10 Oct 2026: the endpoint and install command are unchanged. Do not confuse this with the single-video `twelvelabs` entry, which is a separate server on a separate host with an API key.
Also in Editing & Post
Index, search, and analyze video libraries with Pegasus and Marengo through MCP.
Hosted video-understanding MCP: transcripts, shots, visual scenes, soundbites, and clips.
Search, analyze, and generate edits from source videos through Video Jungle.
Local MCP and CLI toolkit for agent-driven video composition, editing, analysis, and rendering.
Hosted MCP that clips, captions, dubs, reframes, and publishes long-form footage.
Adobe's second hosted MCP — compose and run Firefly Creative Production batch pipelines.
Pairs well with
Quoted dubbing, subtitles, transcription and multi-voice audiobooks from one remote MCP; a human approves each price.
Official Deepgram server: Nova transcription, diarization, and audio intelligence.
Official hosted Airtable MCP: query, create, and update production-tracking bases.