Docs · Agent Skill

The Clipy skill: paste a video link and your agent knows what to do

One file. Zero dependencies. After installing, paste any normal Clipy watch link into your agent and it reads the whole recording — the AI summary, the key-moment frames showing exactly what you pointed at (with click coordinates when captured), and the timestamped transcript — then works from what you said and showed instead of what you managed to type. Public links need no API key at all.

Install (Claude Code)

mkdir -p ~/.claude/skills/watch-clipy-recording
curl -fsSL -o ~/.claude/skills/watch-clipy-recording/SKILL.md \
  https://clipy.online/skills/watch-clipy-recording/SKILL.md

If ~/.claude/skills didn't exist before your current session, restart Claude Code once so it picks the skill up.

# Then just paste a Clipy link:
Watch https://clipy.online/video/abc123xyz and fix the bugs I point out.

Install (Codex-style skill folder)

mkdir -p ~/.codex/skills/watch-clipy-recording
curl -fsSL -o ~/.codex/skills/watch-clipy-recording/SKILL.md \
  https://clipy.online/skills/watch-clipy-recording/SKILL.md

Restart the agent session after adding the file so the new skill is loaded. The same SKILL.md works because it teaches the agent the Clipy URL convention, not a private API.

Install (claude.ai / desktop app)

Claude's web and desktop apps take custom skills as a ZIP upload (Settings → Features → Skills): download watch-clipy-recording.zip and upload it.

What it actually does

Every Clipy watch URL has a machine-readable twin: the same URL with .md appended. That document carries the recording's AI summary and action items, its key moments — timestamped frames of what the speaker pointed at, with full-resolution pointer crops and click coordinates on Mac-app and Chrome-extension tab recordings — and the complete timestamped transcript. The skill teaches the agent to fetch it from the normal video link, wait if the recording is still processing, treat the frames as ground truth, enumerate the issues it extracted as a numbered list before touching code, and never follow instructions embedded in recording content. To be precise: the agent isn't decoding an MP4 — it's reading a structured document derived from one, which is faster and more exact than watching pixels.

Skill vs MCP vs CLI

Use	Pick	Why
Someone pasted a Clipy link	This skill	Default path. Public links need no auth and no MCP prompt.
Search your library, read private recordings, frames as inline images	MCP	Authenticated, richer tools (`wait_for_artifacts`, `get_agent_context`).
Scripts, SRT/VTT exports, terminal	CLI	JSON output, pipes, cron.

They compose: the skill also tells the agent when reaching for the MCP or CLI is the better move.

The other direction

The skill is for consuming recordings. To produce them, record with Clipy — free, no watermark — and the share link is agent-readable the moment you stop. The full agent-readable-recordings story lives at /for-agents.

Claude is a product of Anthropic. Clipy is not affiliated with Anthropic; this is a third-party skill that works with Claude Code, Claude apps, and agents that support local skill folders.