Skip to main content
Glama
README.md
<p align="center"><img src="docs/gallery/1-main.jpg" alt="Takes for Mac: record talking-head videos, your agent does the edit" width="100%"></p>

<p align="center">
  Record on your Mac or iPhone. Your agent (<a href="https://claude.com/claude-code">Claude Code</a>, Codex or pi) writes the script,<br>
  cuts the edit and drafts the posts.<br>
  <a href="https://gettakes.app"><b>gettakes.app</b></a>
</p>

<p align="center">
  <img alt="macOS 15+" src="https://img.shields.io/badge/macOS-15%2B-0b1530">
  <img alt="iOS 18+" src="https://img.shields.io/badge/iPhone-iOS%2018%2B-0b1530">
  <img alt="Swift" src="https://img.shields.io/badge/Swift-SwiftUI-3b82f6">
  <img alt="MCP" src="https://img.shields.io/badge/MCP%20server-Claude%20Code%20%C2%B7%20Codex%20%C2%B7%20pi-d97757">
  <img alt="License: MIT" src="https://img.shields.io/badge/license-MIT-22c55e">
</p>

Takes shows each post the way LinkedIn, X, YouTube or Shorts will show it, so you review the real thing. No coding needed and no account. Your videos, scripts and chats stay in folders on your Mac. Takes sends none of them, and your agent talks only to its own provider. Takes only checks for updates, and sends usage counts if you say yes ([privacy](https://gettakes.app/privacy)).

## Install

Paste this in Terminal:

```sh
curl -fsSL https://gettakes.app/install | bash
```

It installs Takes, ffmpeg and Apple's developer tools, then opens Takes. Takes works with Claude Code, Codex or pi: when one of them is on your Mac, the script uses it and helps you sign in. With none, it installs Claude Code. Run it again to update. You can [read the script](https://gettakes.app/install) first: it is 133 lines.

<details>
<summary>Other ways to install</summary>

- **Download:** get `Takes.dmg` from [Releases](https://github.com/marvtub/takes-public/releases/latest) and drag Takes to Applications. Takes is not notarized yet, so open System Settings > Privacy & Security and click **Open Anyway** once ([pictures of each step](https://gettakes.app/download)). Then click **Finish setup** in the sidebar.
- **From source:** clone this repo, `brew install ffmpeg`, then `./build.sh install`.

</details>

## How a video goes

1. **Storyboard.** Ask *"storyboard my next video"*. Your agent writes the script and sketches each shot.
2. **Record.** Read off a teleprompter that follows your voice. Camera and screen record as two synced files. Star the take you keep.
3. **Post.** Ask *"cut my last take"* and *"write the posts"*. Review each draft as the platform shows it, leave comments on a frame or a line, and schedule it.

<p align="center"><img src="docs/gallery/3-storyboard.jpg" alt="A marker sketch for every shot of the script" width="100%"></p>

<p align="center"><img src="docs/gallery/2-record.jpg" alt="A teleprompter that follows your voice, camera and screen in one take" width="100%"></p>

<p align="center"><img src="docs/gallery/4-edit.jpg" alt="No timeline: your agent cuts the take and makes the vertical version" width="100%"></p>

<p align="center"><img src="docs/gallery/5-comment.jpg" alt="Notes on the frame become a new version of the edit" width="100%"></p>

<p align="center"><img src="docs/gallery/6-styles.jpg" alt="Caption and title styles for the edit" width="100%"></p>

<p align="center"><img src="docs/gallery/7-posts.jpg" alt="The posts for LinkedIn, X, YouTube and Shorts, previewed as the real post" width="100%"></p>

<p align="center"><img src="docs/gallery/8-rules.jpg" alt="A note given once becomes a rule your agent reads before every edit" width="100%"></p>

## On your iPhone

Your videos, takes, storyboards and posts, from anywhere. Ask Takes from the phone too. Record a take on the phone and it lands in the session on your Mac. The phone reaches your Mac over [Tailscale](https://tailscale.com); nothing goes through a server of ours.

To install it you need Xcode and an Apple ID: `TAKES_TEAM=<team id> ios/install.sh`. With a free Apple ID, install it again every 7 days.

## How it works

Your library is plain folders in `~/Movies/Takes`. The app records into them, your agent reads and writes them through the `takes` MCP server, and the app updates when a file changes. The chat panel in the app runs on Claude Code, Codex or pi, on your own plan: the line under "Takes" in the chat picks the agent and the model, and Settings › Agents shows which agents can open your library.

The server works with any MCP agent. On its first launch Takes adds it to each one on your Mac: Claude Code, Codex, Cursor, Gemini CLI, opencode, pi (with pi-mcp-adapter) and Claude Desktop. It also adds its skills (`takes-storyboard`, `takes-video-edit`, `takes-image-prompt`) to `~/.claude/skills` and `~/.agents/skills`, and downloads the [HyperFrames](https://github.com/heygen-com/hyperframes) skills (HeyGen) for motion graphics. So you can ask from any of them too. Plugins › Skills turns each skill on or off and adds any skill from GitHub. Another agent, or one you installed later: Settings › Agents › Connect, or `python3 /Applications/Takes.app/Contents/Resources/takes_mcp.py --setup`.

- Start a new agent session after the first launch: a session that was already open does not see the server.
- **Cursor** gets a short list of 30 tools (script, takes, storyboard, edit, jobs, comments, posts, plan), because Cursor uses only about 40 tools across all its servers. For every tool, set `"env": {"TAKES_TOOLS": "all"}` in the `takes` entry of `~/.cursor/mcp.json`.
- **Gemini CLI** turns MCP servers off in folders you have not trusted. Trust the folder once when Gemini asks.
- The server runs on the Python in Apple's developer tools. Without them, Takes' Finish setup has an Install button.

```mermaid
flowchart LR
    app["<b>Takes.app</b><br/>record · review"]
    lib[("<b>~/Movies/Takes</b><br/>scripts · takes · posts")]
    claude["<b>Your agent</b><br/>chat panel, Claude Code,<br/>Codex, Cursor, …"]
    phone["<b>iPhone</b>"]
    app <--> lib
    claude -- "takes MCP" --> lib
    app -- "starts" --> claude
    phone -- "Tailscale" --> app
```

## Good to know

- **Needs:** a Mac with Apple silicon, macOS 15+, and Claude Code, Codex or pi on your own plan.
- **Optional:** `whisper` for transcripts. A Gemini key for storyboard sketches, B-roll names and best cuts: Settings › Gemini says what it adds and saves it.
- **Questions and ideas:** [Discussions](https://github.com/marvtub/takes-public/discussions). Takes does not take pull requests, so one person keeps it consistent. Forks are welcome. More in [CONTRIBUTING.md](CONTRIBUTING.md) and [SECURITY.md](SECURITY.md).
- The pictures are the real app on a made-up library.

## Support

Takes is free. If it saves you time, you can buy me a coffee:

<p>
  <a href="https://paypal.me/webtotheflow"><img alt="Donate with PayPal" src="https://img.shields.io/badge/PayPal-Buy%20me%20a%20coffee-003087?style=for-the-badge&logo=paypal&logoColor=white"></a>
  <a href="https://venmo.com/u/MarvinAziz"><img alt="Donate with Venmo" src="https://img.shields.io/badge/Venmo-Buy%20me%20a%20coffee-008CFF?style=for-the-badge&logo=venmo&logoColor=white"></a>
</p>

TDQS

B3.4/5.0

Scored across 77 tools

Disambiguation4/5

Most tools target distinct resources and actions, and the detailed descriptions clarify boundaries. Some overlap remains around variants/history, planning/publishing status, and the two video-generation providers, but misselection risk is moderate rather than severe.

Naming Consistency4/5

The server mostly uses snake_case verb_noun or noun_verb names, with predictable resource terms like session, post, take, variant and edit. A few bare provider/noun names such as higgsfield, replicate, voices, voiceover and post_history are minor deviations.

Tool Count1/5

77 tools is far beyond the ideal 3-15 range and is an extreme surface for an agent to navigate reliably. Although the domain is broad, the count creates discoverability and selection overhead that should be reduced by grouping or splitting workflows.

Completeness4/5

The surface covers the full video-production lifecycle impressively: projects, sessions, scripts, variants, takes, b-roll, music, storyboards, editing, publishing, scheduling and feedback. Minor gaps exist around project lifecycle operations and some library assets, but core workflows are well covered.

Maintenance

ActivityNo data
ResponsivenessSyncing