Connect your AI

Direct videos from your own AI app

WiskoCuts is an MCP server with 86 video tools. Connect an AI app once, and it can research, ask for pictures, plan every beat, review the frames and approve the render, in the WiskoCuts app on your computer.

Three ways to connect

What works, and what does not

Supported = we connect it for you and test it with the app. Compatible = a standard MCP app: the installer writes its own documented settings, but it is not in our test runs. Not supported = it cannot reach the app on your computer.

* OpenCode's free models work only inside OpenCode's own chat. To let the app start OpenCode for a step, sign in to a paid model provider in OpenCode. Or direct the video from OpenCode's chat: the tools are connected there either way.

One click in the app

The quickest way, when the WiskoCuts app is installed on this computer.

1. In the app, open Settings → Your AI.
2. Under Connect your AI, press Connect next to your AI app. The app adds its server to that app's settings file. It changes nothing else in the file and saves a copy of the old one as .bak.
3. Fully quit your AI app and open it again, then ask it to start_here.

Claude Code plugin

One plugin installs the MCP server and the Director's how-to (a skill), and Claude Code keeps it up to date.

1. Install uv if you don't have it. It runs the server and fetches its two Python packages the first time.
   WindowsmacOSLinux

   powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

2. Add our plugin marketplace, then install the plugin:

   claude plugin marketplace add https://wiskocut.com/plugins/marketplace.json
   claude plugin install wiskocuts@wiskocuts

   In a Claude Code session, /plugin marketplace add … and /plugin install … do the same.
3. Start a new Claude Code session with the WiskoCuts app open, and ask: start_here.

Plugin 1.0.0+hb5693aabf6 · download the zip (47 KB) · SHA-256 f70a46b98a11e42df641ec7d28d60588b9c4e271e47b945ab9034eeb2a6af18a

Installer for every other app

One download for every app below. It needs Python 3.10 or newer, which you already have if you use a coding AI. Without uv, it installs the server for your own Python instead.

1. Download wiskocuts-mcp.zip (47 KB) and unzip it anywhere.
2. Open a terminal in the unzipped folder and see what is on your computer:

   python install.py list

3. Connect one app, several, or every app it found:

   python install.py cursor
   python install.py cursor gemini opencode
   python install.py all

4. Restart the app(s), keep WiskoCuts open, and ask: start_here.

The installer copies itself to ~/.wiskocuts/mcp/, so you can delete the download, and adds one entry named video-director to the app's own MCP settings. It changes nothing else, keeps the old file as .bak, and writes no key, password or token.

python install.py print vscode    # only show the exact settings text, change nothing
python install.py remove cursor   # take WiskoCuts out of Cursor's settings
python install.py cursor --python # run the server with this Python instead of uvx

Every app, step by step

Each section below gives the installer command, then the settings it writes, for anyone who prefers to do it by hand. In the settings, <HOME> is your home folder (C:/Users/you, /Users/you or /home/you). python install.py print <app> shows the exact text, already filled in.

Claude Code

Use one click or the plugin. To add it by hand instead:

claude mcp add video-director --scope user -e VOX_MCP_AGENT=claude -- uvx --from "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server" wiskocuts-mcp

Check it with claude mcp list: video-director shows as connected while the app is open.

Claude Desktop

python install.py claude-desktop

Or open Claude Desktop's Settings → Developer → Edit Config and add the server to claude_desktop_config.json:

{
  "mcpServers": {
    "video-director": {
      "command": "uvx",
      "args": ["--from", "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server", "wiskocuts-mcp"],
      "env": { "VOX_MCP_AGENT": "claude_desktop" }
    }
  }
}

Quit Claude Desktop fully (from the tray or menu bar, not just the window) and open it again.

Codex

python install.py codex

Or with Codex's own command, or in ~/.codex/config.toml:

codex mcp add video-director --env VOX_MCP_AGENT=codex -- uvx --from "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server" wiskocuts-mcp

[mcp_servers.video-director]
command = "uvx"
args = ["--from", "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server", "wiskocuts-mcp"]
env = { VOX_MCP_AGENT = "codex" }
tool_timeout_sec = 120

OpenCode

python install.py opencode

This adds the server to ~/.config/opencode/opencode.json and the Director's how-to to ~/.config/opencode/skills/video-director/. OpenCode 2.x reads MCP servers from mcp.servers; the installer checks your version and writes the shape it reads. By hand, for OpenCode 2.x:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "servers": {
      "video-director": {
        "type": "local",
        "command": ["uvx", "--from", "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server", "wiskocuts-mcp"],
        "environment": { "VOX_MCP_AGENT": "opencode" },
        "enabled": true
      }
    }
  }
}

Cursor

python install.py cursor

Or add a custom MCP server in Cursor Settings (the MCP section), which edits ~/.cursor/mcp.json:

{
  "mcpServers": {
    "video-director": {
      "command": "uvx",
      "args": ["--from", "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server", "wiskocuts-mcp"],
      "env": { "VOX_MCP_AGENT": "cursor" }
    }
  }
}

Use it from the Agent chat. Cursor lists the server under MCP with a green dot once it has started.

VS Code (GitHub Copilot)

python install.py vscode

Or run MCP: Open User Configuration from the command palette (Ctrl+Shift+P). VS Code uses servers and a type:

{
  "servers": {
    "video-director": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server", "wiskocuts-mcp"],
      "env": { "VOX_MCP_AGENT": "vscode" }
    }
  }
}

The tools show in Copilot Chat's Agent mode, under the tools button.

Windsurf

python install.py windsurf

Or add the same mcpServers entry as for Cursor to ~/.codeium/windsurf/mcp_config.json, with "VOX_MCP_AGENT": "windsurf", then refresh the MCP servers in Cascade.

Gemini CLI

python install.py gemini

Or add the same mcpServers entry to ~/.gemini/settings.json, with "VOX_MCP_AGENT": "gemini". In Gemini CLI, /mcp lists the server and its tools.

Kiro

python install.py kiro

Or add the same mcpServers entry to ~/.kiro/settings/mcp.json.

LM Studio

python install.py lmstudio

Or open the Program tab → Install → Edit mcp.json (~/.lmstudio/mcp.json) and add the same mcpServers entry. Directing a video takes a model that is good with tools: a small local model can struggle with a full plan.

OpenClaw and PicoClaw

These own their settings through their own command line, so the installer shows the exact command instead of writing a file:

python install.py print openclaw
python install.py print picoclaw

OpenClaw

Stop the OpenClaw Gateway first, run the command, then start the Gateway again:

openclaw mcp set video-director '{"command":"uvx","args":["--from","<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server","wiskocuts-mcp"],"env":{"VOX_MCP_AGENT":"openclaw"}}'

PicoClaw

picoclaw mcp add video-director --env VOX_MCP_AGENT=picoclaw -- uvx --from "<HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server" wiskocuts-mcp
picoclaw mcp test video-director

Zed, Cline, Roo Code, JetBrains, Continue, Goose

Each of these takes the standard server entry in its own MCP settings: the command uvx, the arguments --from <HOME>/.wiskocuts/mcp/1.0.0+hb5693aabf6/server wiskocuts-mcp, and the variable VOX_MCP_AGENT set to the app's name.

• Cline / Roo Code: the MCP Servers icon → Configure MCP Servers; paste the mcpServers entry shown for Cursor.
• JetBrains AI Assistant: Settings → Tools → AI Assistant → Model Context Protocol (MCP) → Add → paste the mcpServers JSON.
• Zed: add it under context_servers in Zed's settings, with command, args and env.
• Continue: add a block under mcpServers in its config.yaml, with name, command, args and env.
• Goose: goose configure → Add Extension → Command-line Extension → paste the whole command.

Which stages your AI decides

Your AI never replaces the voice, the layout or the render: those are always the app's. For the stages where a model decides, you choose the app's own models (Gemini on your keys) or your AI over MCP:

• For every video: Settings → Your AI → AI per stage.
• For one video: in the New-video wizard, open More choices, then Directed by → My AI over MCP, or set each step on its own.
• Run my AI from the app: for Claude Code, Codex and OpenCode, the app can start your AI by itself for the one step that is waiting, so you only watch the app. It runs with your own sign-in, sees only the app's video tools (no files, no shell, no web), and every step it takes shows in the video's log.

Your first video from your AI

With the app open, ask your AI app something like:

Use the video-director tools: start_here, then make a 2-minute documentary about the history of the Silk Road.

Your AI calls start_here, creates a draft, and the app opens the New-video wizard for you to check and press Start. From then on your AI does the stages it owns, and waits with wait_for_change until the next one is its turn. The full list of tools is in the MCP tools reference.

Troubleshooting

"401 missing or wrong install token"

The WiskoCuts app is not running, or a different copy of the app is. Open the app, then try again.

"This connection to the app is out of date"

The app was updated while your AI app was open, so new tools are missing. Fully quit your AI app and open it again.

The server does not start: "uvx" not found

Install uv (see the plugin steps), then restart your AI app. Or run the installer with --python, which needs no uv.

My AI app says there are too many tools

Some apps cap how many tools can be on at once. This server offers 86. Turn off the tool groups you don't use in that app's MCP settings; the tools reference shows the groups.

Nothing changed after I connected

Most apps read their MCP settings only when they start. Quit the app fully, not just its window, and open it again.

Security and privacy

• The server and its settings never contain a key, password or token. It asks the running app on this computer for a one-time local pass, and the app answers only on 127.0.0.1.
• Your AI uses its own subscription and sign-in. Your WiskoCuts keys are never sent to it.
• The server can only call the app's video tools: it cannot read or write your files.
• Removing it: python install.py remove <app>, or claude plugin uninstall wiskocuts.