Fast Video Cataloger 10.3 adds AI assistants and trained tags

Fast Video Cataloger has launched version 10.3 for Windows, linking assistants such as Claude to local video libraries through the Model Context Protocol and adding a new Trained Tags feature. The update is aimed at making large on-device archives easier to search, organize and analyze without moving media to the cloud.

Why it matters: - Fast Video Cataloger 10.3 gives editors, archivists and content teams a new way to search local footage libraries in plain language. - The update keeps video files and catalogs on users’ own machines while adding AI-assisted search, inspection and organization. - The release targets large archives where transcripts, cast data and scene metadata have accumulated over years.

What happened: - Fast Video Cataloger released version 10.3 of its Windows video cataloging software. - The update connects AI assistants such as Claude to self-hosted video libraries through the Model Context Protocol, or MCP. - The connector, fvc-mcp, is included in the installer and works with the Fast Video Cataloger server. - A user can ask an assistant for footage in plain language, and the assistant can search the catalog, inspect scene thumbnails, and organize results.

The details: - The assistant can search videos, scenes, transcripts and cast. - The assistant can visually inspect scene thumbnails to verify what is in a shot. - The assistant can apply and remove keywords. - The assistant can gather clips into bins from a single request. - Server API keys now support long-lived credentials with separate permission levels, including read-only access. - Administrators can create and revoke API keys in the User Management window or through the REST API. - Only a fingerprint of each key is stored. - The number of active keys follows the number of user accounts allowed by the license. - Videos already in a catalog can now be indexed through the REST API. - That lets a script or assistant process a backlog on its own, including transcription and AI analysis. - Trained Tags lets a user teach a custom keyword by selecting about ten thumbnails that show a subject and choosing Teach trained tag. - The software then finds that subject across the catalog and adds the tag automatically during indexing or in a batch. - A tag can match whole frames or attach to a detected object. - A wrong match can be corrected with a right-click that also teaches the tag. - No dataset annotation or training run is required. - AI image tagging from version 10 can now run while videos are being indexed, so new material can arrive already tagged. - Tagging now calibrates itself per catalog, so the software learns what is typical for that library. - That makes tagging more selective for libraries dominated by screen recordings or interviews.

Between the lines: - The MCP integration turns the video catalog into a tool an assistant can operate directly, instead of treating AI as a separate search layer. - The local-first setup is aimed at users who want AI assistance without sending footage or metadata to a cloud service. - Trained Tags narrows the gap between generic AI keywords and the specific subjects inside a user’s own archive. - The per-catalog calibration suggests the software is adapting to different library types instead of using one fixed tagging threshold.

What's next: - Fast Video Cataloger 10.3 is available now for Windows 10 and 11. - Pricing starts at $9.90 per month, $97 per year, or $197 for a perpetual license. - A free 30-day trial is available at videocataloger.com. - The company says the full list of fixes and refinements is in the version history in the user guide.

The bottom line: - Fast Video Cataloger is betting that local video libraries, not cloud storage, will be the most useful place for AI search and tagging to live.