VidXP plugin
Local video search: speech, sounds, scenes and actions found by meaning, with frames and clips returned to agents.
- 59 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Indexes video recordings, including faces, on your machine and downloads several gigabytes of models; server mode exposes an HTTP API.
Install
- Works in
Manual install
Add https://github.com/grayhatdevelopers/vidxp as a Git plugin marketplace, install the VidXP plugin, then use its $vidxp-install skill to set up VidXP on this computer.
Paste this prompt into Codex; the plugin brings an install skill and two video skills.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Indexes video recordings, including faces, on your machine and downloads several gigabytes of models; server mode exposes an HTTP API.
This is third-party code. Review the repository files before installing.
What it does
VidXP makes one video or a whole collection searchable by meaning. Speech search finds a moment from a paraphrase, scene and action search take a description of what is on screen or happens over several seconds, sound search works on events, and actor matching groups recurring people within a video and exports a highlighted clip. For agents VidXP ships as a Codex plugin with three skills (install and connect, ingest and index, find moments with evidence) and as an MCP server through which a client adds videos, searches the library and returns frames, boards and clips. `vidxp mcp-config` prints the settings for a local client. A desktop app, CLI, browser interface, HTTP API and a Premiere Pro extension are included.
Who it is for. Editors, archivists and teams holding years of video who want to ask an agent where something was said or appeared.
Good fit when
- In a family archive you need a moment from a phrase you remember loosely.
- An editor must pull scenes with a given action out of hours of footage.
- An agent answers a question about a video library and returns frames as evidence.
- Video stays on your own machine and cloud search services are not an option.
Not a fit when
- You need a one-off look at a short clip with no indexing.
- The computer has no room for models: search downloads about 1 to 3 GB.
- People faces must not be processed even locally.
Example request
Find the moment in my videos where someone talks about renewing the contract and show frames and a clip of it.Limitations
Windows, Apple Silicon macOS and Linux are supported. The Python runtime the desktop app manages takes about 3 GiB, and models download per chosen capability: speech 2.64 GiB, scenes 1.43 GiB, sound 0.94 GiB, actions 0.93 GiB, faces 37 MiB. Command-line installation needs `uv`, Python 3.14 and FFmpeg. Faces are grouped within a single video and no identity is established from them. Server mode and the HTTP API are set up separately per the installation guide, and the Premiere Pro extension installs independently of the Codex plugin.
How to disable. Remove the VidXP plugin from the Codex marketplace, run `uv tool uninstall vidxp` and delete the directory with the index and downloaded models.
FAQ
How do I connect VidXP to Codex?
Paste a request into Codex to add the `grayhatdevelopers/vidxp` repository as a Git marketplace, install the plugin and call the `vidxp-install` skill. The desktop app does the same with its Set up in Codex button.
How do I install from the command line?
Install `uv`, run `uv tool install --python 3.14 --torch-backend cpu "vidxp[local-worker,mcp]"`, then `vidxp init`, `vidxp prepare` and `vidxp doctor`. These check FFmpeg and download models.
Which kinds of search exist?
By speech, by sound events, by scenes, by actions and by recurring faces within a video.
Do videos go to the cloud?
By default no: video stays on your machine. Hosted clients can connect to a deployed VidXP server if you run one.
How do I open the browser interface?
Install the package with the `frontend` extra and run `vidxp ui`.
Discussion
No questions or reviews yet.
Related
PluginsMCP server, /video plugin and CLI: transcript, key frames, on-screen text and metadata for a YouTube, Loom or local video
VideoDB skills and an MCP server teach an agent to upload video, cut clips, add subtitles and find moments by meaning
Reelpack breaks a local video into transcript, cuts, on-screen text and music beats so an agent reads it without watching.
The eyes CLI and skill for agents: a video index, search by words and on-screen text, close-up frames and editing metrics
Cloudglue MCP: video analysis for agents
Cloudglue MCP Server
An agent in Cursor or Claude Desktop describes video, splits it into chapters and shots and finds moments by speech or picture
Claude Video Vision
claude-video-vision
An MCP server and Claude Code plugin that let the agent watch video: frames via ffmpeg and audio transcription via Gemini, local Whisper or OpenAI
README in short
The project is presented as a local search engine for people and AI agents with a desktop app, CLI, MCP and HTTP API. Models download per capability, with size and destination shown first.