Give your AI coworker specialist video tools

Overview

Cadre is specialist recording and non-destructive video-editing software for Apple Silicon Macs. Its local Agent API lets a compatible desktop assistant operate the same project and timeline the person sees in the app.

Cadre's editor is fully data-driven — a project is a manifest plus timeline data, and the app already speaks a typed command protocol internally. The Agent API wraps that surface in a Model Context Protocol (MCP) server that runs inside the app, bound to 127.0.0.1 on an ephemeral port.

Recording uses the same main-process start path rather than a privileged shortcut: an agent can start, pause, resume and stop a take, and Cadre creates the floating recording controls. The countdown defaults on but can be disabled; screen/area highlighting and display dimming are conditional cues. Everything after the take — inspecting the timeline, zooming, cutting, captioning, styling and exporting — is available as tools your agent can call.

Read the complete open specification

The concise guide on this page is backed by three complete, server-rendered references. Each page also links its exact Markdown encoding for tools that prefer plain text.

• Agent API overview — discovery, MCP connection, authentication, lifecycle, security and errors.
• 62-tool reference — every parameter, constraint, result, error and worked workflow.
• Project format — the .screencraft layout, persisted fields and live-versus-offline editing boundary.
• Public cross-agent skills and Claude plugin source — independently versioned standalone v1.1.0 with the MIT-licensed editing and motion-graphics skills plus the token-free local MCP bridge. Its dated compatibility record covers Cadre 1.0.0-rc.33's asynchronous 60-tool contract and background-job lifecycle; it is historical evidence, not the current release identifier or primary Work/Cowork setup guidance.
• Current compatibility evidence — authenticated rc.33 and standalone plugin v1.0.2 MCP initialization, exact ordered 60-tool discovery and one benign get_app_state call, with explicit non-proofs.
• Earlier rc.20 compatibility evidence — the frozen record of the same plugin against rc.20's 57-tool surface, superseded by the rc.33 record above.
• Measured handshake evidence — the historical signed rc.18 build's reproducible 54-tool handshake and sanitized local latency record.

These documents describe the public local control surface. They do not expose Cadre's private application source code or create a hosted remote API.

Cadre Agent API at a glance

Connect your desktop assistant

Install the ChatGPT or Claude desktop app on the same Mac as Cadre. Open a recording, choose Edit with AI, and write the result you want in What would you like to change?. Cadre includes the selected project ID and your brief in the request it prepares.

● Recording open
● Project included in the draft
● Request remains yours to send

› Turn this into a concise customer tutorial.
  Keep the setup explanation, cut the pauses, focus the
  product and add clear captions.

ChatGPT Work

Click Connect & open. Cadre installs its local plugin and opens a ChatGPT draft with your brief and selected project. Review the draft, press Send, and keep ChatGPT and Cadre open while the assistant works. Later requests use Open in ChatGPT. Cadre refreshes the private local connection after restarts.

Claude Cowork

Click Set up Cowork. Cadre prepares a ZIP containing its editing and motion-graphics skills plus a token-free local bridge; no Node.js installation or terminal command is required. Upload that ZIP once in Claude's Customize → Plugins screen and enable Cadre. Then return to Cadre, choose Open in Cowork, review the prepared draft and press Send. Keep Claude Desktop and Cadre open while editing.

Cadre sells the recorder, editor, renderer and integrations. It includes no AI credits or Cadre-funded inference. Assistant usage comes from your own ChatGPT or Claude account and remains subject to that provider's plan, limits, permissions and data policy.

Auth & security

• Localhost only. The server binds to 127.0.0.1 on an ephemeral port — never a public interface.
• Bearer token on every request. A fresh 32-byte token is minted each boot; supplied and expected values are SHA-256 hashed, then the fixed-width digests are compared with timingSafeEqual.
• Owned long jobs. Imports, transcription and model downloads return immediately with a separate random job token. That exact token is required to poll or cancel, results expire after 15 minutes, and all job state disappears on restart.
• Recording uses the shared UI path. An agent can start, pause, resume and stop a recording; Cadre creates the floating recording controls. The countdown defaults on but can be disabled, while screen/area highlighting and display dimming are conditional cues. macOS TCC still governs desktop capture.
• Nothing you type is recorded. Cadre logs clicks, scrolls and shortcut combinations — never the characters you type. There is no typed text in a project for an agent to read, because none was ever written.
• The paywall still applies. Export is licence-gated server-side — an agent hits the same LICENSE_REQUIRED gate the UI does, and cannot bypass it.

Tool reference (v1)

The v1 server exposes 62 tools. Calls validate their arguments and return verifiable results. Imports, transcription, caption generation and model downloads return a job receipt immediately, then use explicit status/result tools so a client's short call deadline cannot interrupt valid local work.

Example prompts

You talk to your agent the way you'd talk to an editor. A few that map cleanly onto the tools above:

Errors

Failures come back as a structured { code, message, hint } object, so an agent can branch on the code rather than parse prose.