Connecting an AI agent
Clip Cabinet speaks MCP (Model Context Protocol), so an AI agent can search and read your whole clip library on your behalf. Works with Claude Code, Cursor, Claude Desktop, and any MCP-aware client.
Connect
- Mint a token at /settings/tokens.
- Paste the Clip Cabinet MCP config into your client and add the token.
- Your agent can now work with your clips.
The config snippet and full setup steps are on the MCP reference at /docs/mcp.
What your agent can do
Once connected, the agent gets these tools:
- list_recordings: your recent clips, newest first.
- search_recordings: find clips across your library by meaning, not just keywords.
- get_recording: full detail for one clip, including every pinned comment and the exact frame image each one points to.
- search_within_recording: find a moment inside a clip.
- get_frame: a captioned frame image.
- get_audio_clip: a slice of a clip's audio.
- list_artifacts: the URLs, errors, and code pulled from your clips.
Comments are the strongest signal
A comment pinned to an exact spot and moment is a precise note — "change this text to that," "this shouldn't be here." Agents treat pinned comments as the authoritative issue list for a clip, above whatever they'd infer from the transcript alone, and can see the exact frame each one is pinned to rather than just its caption.
Security
The agent authenticates with a bearer token. Tokens are hashed at rest and you can revoke any of them at /settings/tokens. See API tokens.