# Agents · VideoDB

> Add VideoDB to your agent.

Paste a link or upload a file, then ask. Your agent finds the moment, cuts the clip and sends back a link you can play. Works in Claude, Claude Code, Cursor or any MCP client. $20 of free credits, no credit card.

Canonical: https://videodb.io/agents · Add to Claude: https://videodb.io/claude · MCP server: https://mcp.videodb.io/mcp

## Install

Every client connects to the same server. Log in to VideoDB once and the tools are there. Steps per client: https://docs.videodb.io/pages/mcp/connect

- **Claude** (web, desktop, mobile): add VideoDB from the plugin directory (https://videodb.io/claude), log in or sign up when VideoDB opens, click Authorize. To add it by hand, open Settings, then Connectors, then Add custom connector, paste `https://mcp.videodb.io/mcp` and click Connect. On Team and Enterprise plans an admin adds it under Organization settings.
- **Claude Code**: `/plugin marketplace add video-db/skills`, `/plugin install videodb@videodb-skills`, `/videodb setup`. Or `claude mcp add --transport http videodb https://mcp.videodb.io/mcp`, then `/mcp` to authorize.
- **Cursor**: Settings, Cursor Settings, MCP, Add new global MCP server, with `{"mcpServers": {"videodb": {"url": "https://mcp.videodb.io/mcp"}}}`. Cursor opens the VideoDB login on first use.
- **Any coding agent** (Codex, Copilot and the rest): `npx skills add video-db/skills`, then ask the agent to set up VideoDB with an API key from https://console.videodb.io.
- **Any MCP client**: add `https://mcp.videodb.io/mcp` as a remote server over Streamable HTTP. Log in once. Ask it to list your VideoDB collections to confirm.
- **Code**: `pip install videodb` or `npm install videodb`. Docs: https://docs.videodb.io

## What people ask, and what comes back

Every answer comes with the times and frames behind it.

- "Make chapters and notes for this lecture." You get timestamps, a chapter list and a summary you can paste anywhere.
- "Find every moment the product is on screen." You get the clips, each one playable, with the frames that matched.
- "Pull every slide from this conference talk." You get the slides as images, with the time each one appeared.
- "Review this match and list the moments to study." You get the moments as clips, with a note on each.
- "Subtitle this video in English, French and German." You get subtitle files in every language, from one transcript.
- "Find where pricing comes up in these sales calls." You get the moments from each call as clips, with what was said around them.

## What your agent can do

70 tools in six groups. You ask for the result. The agent picks the tools; these are the ones it reaches for most.

- **Bring video in**: files, links, YouTube, buckets, live cameras, screens and microphones. `upload_media`, `connect_rtstream`, `transcode_media`.
- **Understand it**: describe scenes, transcribe speech, read on-screen text, return your own JSON fields. `understand_video`, `generate_transcript`, `create_index`.
- **Search it**: plain-language search over one video or a whole collection. `search_video`, `search_collection`, `ask_video`.
- **Edit it**: clips, timelines, vertical reframing, subtitles, voice and music. `compile_clips`, `reframe_video`, `add_subtitles`.
- **Stream any result**: every answer as a playable link, for a person or the next tool. `generate_clip`, `render_timeline`, `get_media_download_url`.
- **Watch live**: connect cameras, set alerts in plain words, get the clip behind each one. `create_rtstream_alert`, `search_rtstream`, `understand_rtstream`.

The layers underneath: https://videodb.io/products.md

## Build your own agent from open code

A repo or a write-up to start from, for each kind of agent. https://github.com/video-db

- Archive agents: find, cut and caption moments in games, streams, podcasts and meetings. Start from Director: https://github.com/video-db/Director (more: https://videodb.io/archival-footage.md)
- Computer-use agents: capture the screen, understand each step, turn a recording into a skill and keep every run on record. Start from open-record-replay: https://github.com/video-db/open-record-replay (more: https://videodb.io/computer-use.md)
- Physical agents: memory for physical devices with cameras. Start from How a robot remembers: https://videodb.io/blog/how-a-robot-remembers.md
- Research agents: search datasets and test runs in plain language, count what is there and export a training set. Start from https://datasets.bot (more: https://videodb.io/data-providers.md)
- Fleet agents: find any scenario across every drive, flight and camera. Start from https://videodb.io/autonomous-vehicles.md

## Questions

- **Is it free to start?** Yes. New accounts start with free credits and no credit card. After that you pay for storage, processing and search.
- **Where do my videos go?** Into your own VideoDB collection, which only your account and your agents can reach. On your own cloud, the files stay in your bucket. Deleting a video removes it in full.
- **Which clients work?** Claude on web, desktop and mobile, Claude Code, Cursor, and any client that speaks MCP over Streamable HTTP. Python and Node SDKs for code.
- **Can it handle long videos?** Yes. An hour-long recording or a day of camera footage is searched the same way, and the agent reads only the moments it needs.
- **Can I bring my own model?** Yes. Any VLM, speech or detection model runs as an analyzer: open-weight, proprietary or your own.
- **Do I need to code?** No. In Claude you add the plugin and ask. The SDKs and the skills are there when you want to build the agent yourself.

## Add VideoDB to Claude

It takes about a minute. Then paste a link or upload a file and ask. [Add to Claude](https://videodb.io/claude) · [Talk to us](https://videodb.io/contact.md)