https://github.com/user-attachments/assets/27b977aa-65b6-4704-99f5-a04dd79f8c7e
Watch the full demo on YouTube
| Skill | What the agent can do | Needs |
|---|
video-player | Build OTIO timelines from local files or URLs with cuts, layers, titles, effects and music. Preview on localhost, edit alongside the assistant in a full-featured editable timeline, export MP4 to the project folder, or move edits between your machine and Tellers with the CLI. | Python 3 + a browser (Chrome / Edge recommended). No Tellers account. |
tellers-mcp | Install the Tellers MCP, edit cloud projects, search footage and transcripts, generate voice, music, images and video, and open a live preview. | A Tellers account (create one on the sign-in page). |
Both use the same timeline format, documented in
otio-format.md.
Edit alongside your assistant
The video-player skill includes the full-featured Tellers editable timeline,
the same one used in the Tellers app. Ask your assistant to show the timeline to
trim, split and move clips, edit titles, and adjust effects and transitions while
the preview updates.
Add markers and notes at precise locations in the timeline to guide the next
edit. Press M to add a marker at the playhead, even during playback, then give
it a name and note. Your edits and feedback are saved locally in OTIO so the
assistant can pick them up and continue from your latest version.
To open it yourself, add &timeline=1 to the local player URL:
http://localhost:8000/player.html?otio=timeline.otio&timeline=1.
Install
Use the skills CLI as an alternative to the native Claude Code and Codex plugin commands below:
npx skills add tellers-ai/skills
Claude Code (plugins)
claude plugin marketplace add tellers-ai/skills
claude plugin install video-player@tellers
claude plugin install mcp@tellers
Or from inside Claude Code: /plugin marketplace add tellers-ai/skills, then
/plugin. The mcp plugin also registers the Tellers MCP server; run /mcp →
tellers → Authenticate once to sign in.
No Tellers account yet? The sign-in page has Continue with Google and a Create
account tab. With an email and password, open the confirmation link Tellers emails
you; the page then unlocks Approve and continue on its own. tellers login
(Tellers CLI) opens the same page.
Codex
codex plugin marketplace add tellers-ai/skills
codex plugin add video-player@tellers
codex plugin add mcp@tellers
Or install them from the Plugins directory in the Codex app after adding the
marketplace. The mcp plugin registers the Tellers MCP server; sign in with
codex mcp login tellers. Codex reads .agents/plugins/marketplace.json and each
plugin's .codex-plugin/plugin.json; Claude Code reads the .claude-plugin/ files.
Both use the same skills/ folders.
Any agent that reads skill folders
Copy a skill folder into your skills directory: ~/.claude/skills/ for Claude Code,
~/.agents/skills/ for Codex (or .agents/skills/ inside a repo):
cp -R plugins/video-player/skills/video-player ~/.agents/skills/
cp -R plugins/mcp/skills/tellers-mcp ~/.agents/skills/
For claude.ai / Claude Desktop, zip a skill folder and upload it under
Settings → Capabilities → Skills.
Layout
.claude-plugin/marketplace.json Claude Code marketplace
.agents/plugins/marketplace.json Codex marketplace
plugins/
video-player/
skills/video-player/
SKILL.md
references/otio-format.md timeline format
references/transfer-to-tellers.md local edit -> Tellers project with the CLI
assets/player.html full-page player (bundle.js from tellers.ai/videoplayer)
assets/example.otio
scripts/serve.py localhost server with Range + CORS; saves exports
scripts/check_otio.py timeline and media URL checker
scripts/prepare_transfer.py upload folder for recreating the edit with the Tellers CLI
scripts/pull_from_tellers.py Tellers project -> local media + .otio
mcp/
.mcp.json registers https://mcp.tellers.ai/ (both agents)
skills/tellers-mcp/SKILL.md
skills/tellers-mcp/agents/openai.yaml Codex metadata + MCP dependency
The player
The player is bundle.js (TellersPlayerUI), the same player that ships in
the Tellers app, published at https://tellers.ai/videoplayer/bundle.js.
Workers, codecs and assets are inlined, so pages on any origin load it with one
<script> tag. player.html?src=<url> tries another build. Playback, export and
subject detection run in the browser, so they depend on that machine; detection
(detect, outline masks) needs a GPU to run at a usable speed.
License
MIT. See LICENSE.
