Control your Mac screen recorder from an AI agent

Joy Truepath · 30 June 2026

You're in the middle of a conversation with Claude or Cursor, debugging something, and you want a screen capture of the repro. Normally that's a context switch: leave the agent, find the recorder, fiddle with it, come back. What if you could just say "grab a ten-second clip of this" and have it happen?

That's the idea behind the MCP server that ships inside TruePath Recorder. Once it's wired up, your agent can start and stop recordings, see what's capturable, and get the saved file path back, without you touching the app. Everything runs on your Mac. Nothing about the recording goes anywhere near a server.

What MCP is, briefly

The Model Context Protocol is a standard way for AI hosts to talk to outside tools. An MCP "server" exposes a few actions; the host (Claude Desktop, Claude Code, Codex, Cursor, and others) can call them mid-conversation. TruePath Recorder ships one of these servers as a small local binary, so "the tool the agent can use" is your actual screen recorder.

It's a thin bridge, and a local one. The server only does what the app's own buttons do, and it never uploads anything.

What the agent can actually do

Four actions, kept deliberately small:

• Check status — whether it's idle, recording, or paused, and how long it's been going.
• List sources — the displays and windows it can capture, with ids, so the agent can target a specific window instead of the whole screen.
• Start recording — with the options you'd expect: display, window, or a region; mic on or off; system audio on or off; H.264 or HEVC.
• Stop recording — which hands back the saved file path, plus the duration and size.

The last one is the quietly useful part. Because the agent gets the path, the recording becomes a step in a larger flow: capture a window, get the file, then do whatever's next in the same conversation.

Setting it up

The server is already inside the app, so there's nothing extra to install.

1. Open TruePath Recorder and go to Settings → AI Agent Control. It's off by default. Turn it on. That starts a local, token-protected control server the MCP talks to.
2. Next to your agent, click Copy. It copies a config snippet with the bundled binary path already filled in.
3. Paste it into your host's MCP config and restart the host.

For Claude Code the whole thing is one line:

claude mcp add truepath -- "/Applications/TruePath Recorder.app/Contents/MacOS/truepath-mcp"

Then just ask, in plain language:

Record a 10-second screen clip with TruePath.

The agent calls start_recording, waits, calls stop_recording, and tells you where the file landed.

The privacy part, because it matters here

Handing a recorder to an AI agent is exactly the kind of thing that should make you slightly nervous, so the design leans the cautious way. It's off until you turn it on, and you opt in per agent. The control server listens on 127.0.0.1 only and needs a token that's regenerated every launch, so nothing off your machine can poke it. And the server's reach stops at start and stop. It can't read your files, and it can't upload a recording, because moving recordings around isn't something it's able to do.

Why bother

If you live in an agent already, the win is just fewer context switches: the recorder becomes one more thing the assistant can reach without you leaving the conversation. Capture a bug as it happens. Record a window while a script runs against it. Get a quick clip mid-task and keep going. None of it is dramatic on its own, which is sort of the point.

A few things people ask

Which agents work with it?
Any MCP client. There are ready-made snippets for Claude Desktop, Claude Code, Codex, and Cursor.

Does anything get uploaded?
No. The bridge is local, token-protected, and only starts and stops the same capture the app's UI does.

Is there a separate thing to install?
No. The server is bundled in the app. Turn on AI Agent Control, copy the snippet, paste it into your agent. The source is on GitHub if you want to build it yourself.