VidXP: MCP tool for Hermes Agent

Local-first video search by speech, sound, scene, action and faces, with MCP access for agents

What VidXP does

VidXP (Video eXPlain) makes one video or a whole collection searchable by meaning. It supports speech search, sound search, scene search, action search for events that unfold over several seconds, and actor matching that groups recurring faces and can render a highlighted video for a selected group. Videos can stay on your own machine, and personal, client or project libraries can be kept separate.

It can be used through a desktop app for Windows, Apple Silicon macOS and Linux, a browser interface, a CLI, an MCP server, an HTTP API, or an all-in-one Docker image for a home server. The repository description presents it as a way to connect Hermes, Claude or OpenClaw agents so they can answer questions with evidence from your own videos, and the CLI prints settings for a local MCP client. Grounded answers come with supporting boards, frames or clips. A Coolify guide covers deployment with separate control and worker images.

Key features

Speech, sound, scene and action search by natural-language description
Actor grouping with highlighted overlay export
Desktop app, browser interface, CLI, MCP server and HTTP API
Grounded questions with supporting boards, frames or clips
Separate libraries for personal, client or project videos
Docker images and a Coolify deployment guide

When to use it

Searching years of family videos by something someone said
Adding video search to an editing workflow
Letting an AI agent answer questions with evidence from your own video library

Who it is for: People, developers and agent users who need to search their own video collections by meaning.

How it fits with Hermes Agent

VidXP is a general video search tool that exposes MCP and an HTTP API, and its repository description names Hermes, Claude and OpenClaw agents as clients.

How to install VidXP

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

uv tool install --python 3.14 --torch-backend cpu "vidxp[local-worker,mcp]"
vidxp init
vidxp prepare
vidxp doctor

Requirements: uv and FFmpeg, with Python 3.14 as used in the install command. Runs on Windows, Apple Silicon macOS and Linux.

FAQ

What is VidXP?

VidXP is a local-first video search engine. It indexes videos so you can find moments by speech, sound, on-screen scenes, actions and recurring faces.

Does VidXP work with Hermes Agent?

The repository description says you can connect Hermes, Claude or OpenClaw agents to it. The README provides an MCP server, and vidxp mcp-config prints the settings for a local MCP client.

How do I install VidXP?

Install uv, then run uv tool install with the vidxp[local-worker,mcp] package, followed by vidxp init, vidxp prepare and vidxp doctor. A desktop installer and a Docker image are also available.