LM Studio
How to let a local model watch a video
LM Studio supports MCP servers. The install button adds Playhead to its list, and any vision-capable model you have loaded can then be handed contact sheets to look at.
Free tier: 60 credits a month, no card. One credit is about a minute of video.
What this changes
The split here is the interesting part: the model stays on your machine and the video never becomes a prompt sent to a hosted model. What crosses the network is the video, and what comes back is images — the reasoning about them happens locally. For a vision-capable local model that is the whole missing piece, because decoding video is the part a laptop is worst at doing on demand.
Setting it up
- 01
Click the install button
It opens LM Studio and adds the server to its MCP list.
- 02
Load a vision-capable model
The sheets come back as images, so the model has to be able to look at one.
- 03
Ask about a video
A URL or a local path.
Things worth asking it
Each of these produces an answer with a second attached, because the timestamp is burned into the frame the model is looking at — it cannot lose track of when something happened.
- Watch this clip and describe what changes between the first and last frame.
- How many distinct scenes are in this video?
How closely it looks is a choice
Six levels, picked by the agent from the question. Surveying an hour and reading a single frame of animation are different jobs, and charging the same for both would make one of them pointless.
| Level | Rate | Answers |
|---|---|---|
| overview | 0.5–2 fps | What is this video, start to end |
| standard | 5 fps | What happens when — cuts, overlays, entrances |
| detail | 5 fps | Objects and text blocks, tiles four times larger |
| fine | 10 fps | Type, logos, subtle transitions |
| motion | 30 fps | Animation, easing, dropped frames |
| single | one frame | Reading text at full resolution |
Questions
- Does the video leave my machine?
- For a URL, we fetch it — it never touches your machine at all. For a local file, it is uploaded to be decoded, because decoding is what the server does. If nothing may leave your network, the Enterprise deployment runs the whole engine inside it.
Or somewhere else
One account, every client. The connection is attached to you rather than to a machine, so adding a second one costs nothing.
- ClaudeWeb, desktop and mobile. Opens the connector dialog with the URL filled in.
Claude CodeOne command in the terminal, then /mcp to sign in.
claude mcp add --transport http playhead https://mcp.tryplayhead.com/mcpNo deeplink exists for the terminal, so this is the install.
- ChatGPTNeeds Developer mode under Settings → Connectors → Advanced first.
- CursorAdds the server to Cursor's MCP settings.
- VS CodeOpens VS Code and adds the server to your MCP configuration.
CodexAdd the server to Codex's config file.
[mcp_servers.playhead] url = "https://mcp.tryplayhead.com/mcp" auth = "oauth"Append to ~/.codex/config.toml — the CLI, IDE extension and desktop app share it.