Skip to main content
Glama
ZermatAbdenour

Gameplay Studio MCP Server

Gameplay Studio (v0.7.0)

Multi-Track NLE Video Editor That Cuts Gameplay From Its Own Label Track • Agent-Driven Over MCP • Bring-Your-Own AI Model & Base URL • Cinema 3D .cube LUTs & Split-Wipe Scopes • Hardware GPU Render

Labelled footage MCP Ready Tests Codecs

📖 Labelled gameplay: how an agent cuts from a label track


Visual Studio Showcase


Related MCP server: mcpCut

Architecture Overview


Highlights in v0.7.0

  • Native Windows Desktop App (GameplayStudio_installer.exe): 4-Zone Resizable Studio Workspace, universal Ctrl+K instant search, hardware-accelerated FFmpeg (NVENC / AMF / QSV / ProRes / HEVC), in-place NSIS Update installer, and 1-click 🔄 Update self-updater.

  • Interactive Before / After Split-Wipe & Broadcast Scopes: Compare raw vs. color-graded video directly inside the Player monitor (⇆ Split Wipe) alongside live RGB Histograms & IRE Waveforms (📊 Scopes).

  • Bring Your Own API Key, Base URL & AI Model (⚙ API & Model): Connect OpenAI, OpenRouter, DeepSeek, NVIDIA NIM, Groq, Ollama, LM Studio, or any custom OpenAI-compatible /v1 endpoint with live /models discovery, connection testing, and natural-language AI Copilot execution on Windows.


Labelled gameplay footage

This is what Gameplay Studio adds on top of a conventional NLE.

A game recorded with the Labeled Recorder Unity package writes a sidecar next to its video:

Recordings/arena_run_003.mp4            the footage
Recordings/arena_run_003.labels.json    what happened in it, frame-accurately

The sidecar lists the events the game fired, the things that lasted, the moments a developer flagged as worth keeping, the chapter structure, and game-data signals sampled whenever they changed. So an agent does not watch the video or guess from pixels — it is told, to the frame, and can justify every cut it makes:

marker 'highlight' (note=triple kill from the rafters) at 00:10.666

labels.attach      bring a labelled recording in (video path, label path, or asset name)
labels.describe    this game's event names, signals, chapters, busiest stretch
labels.find_windows ranked candidate clip ranges, each with a reason
labels.build_edit  put the chosen ranges on a new track, as one undo step
labels.auto_short  do all of the above with sensible defaults

Full workflow, including the strategies for shorts vs. trailers: docs/labelled-gameplay.md.

Unlabelled footage still works; it just falls back to the pixel-based tools (edit.highlight, analysis.scenes, analysis.beats), which guess.


Install

Build from source. Node 20+ and Git are the only requirements; FFmpeg and FFprobe are fetched as npm dependencies.

npm ci
npm run build

Download the Windows installer from Releases, or build it yourself with npm run dist:desktop.


Quick Start (Development & Building From Source)

npm ci
npm test                        # Run all end-to-end Vitest tests
npm run build                   # Build TypeScript backend + CLI + MCP
npm run dev:api                 # Start local Gameplay Studio Engine on http://127.0.0.1:47000
npm run dev:ui                  # Start React + Vite Studio UI on http://localhost:5173

Connect an agent (MCP)

Gameplay Studio is an MCP server, so any MCP-capable agent drives the whole editor — the ~110 ordinary editing tools plus the labels.* family. The first call an agent should make is labels.guide, which returns the entire labelled-footage workflow so it never has to be briefed out of band.

Claude Code

claude mcp add gameplaystudio   --env GAMEPLAY_STUDIO_PROJECT=.gameplaystudio/project.gameplaystudio   -- node "/path/to/Gameplay Studio/dist/mcp/stdio.js"

opencode (any model — DeepSeek, Qwen, local Ollama …) in opencode.json:

{
  "mcp": {
    "gameplaystudio": {
      "type": "local",
      "command": ["node", "/path/to/Gameplay Studio/dist/mcp/stdio.js"],
      "environment": { "GAMEPLAY_STUDIO_PROJECT": ".gameplaystudio/project.gameplaystudio" },
      "enabled": true
    }
  }
}

Claude Desktop / Cursor / anything else speaking MCP

{
  "mcpServers": {
    "gameplaystudio": {
      "command": "node",
      "args": ["/path/to/Gameplay Studio/dist/mcp/stdio.js"],
      "env": { "GAMEPLAY_STUDIO_PROJECT": ".gameplaystudio/project.gameplaystudio" }
    }
  }
}

GAMEPLAY_STUDIO_PROJECT is the .gameplaystudio project file the agent edits, created on first run if missing. GAMEPLAY_STUDIO_PROJECT is still honoured so an existing config keeps working. An HTTP transport is available via gameplay-studio mcp --http.


Credits

Gameplay Studio is a renamed derivative of Vedit Studio by lilrepaer-glitch, used under the MIT licence. The NLE engine, render pipeline, and desktop shell, and the original MCP server are its work. Gameplay Studio adds the labelled-gameplay layer (src/labels, src/mcp/labels.ts) and the rename.

The label format it reads is produced by the Labeled Recorder Unity package.

License

MIT — see LICENSE.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables AI agents and MCP clients to programmatically edit video projects on a local desktop editor, with 119 tools for multitrack editing, effects, captions, audio, and batch auto-editing, producing reviewable and reversible real timeline edits.
    AGPL 3.0
  • A
    license
    Not graded
    quality
    C
    maintenance
    An MCP server that gives AI agents a stateful multi-track video editor powered by MLT, enabling them to assemble timelines, add media and text, apply filters and transforms, preview downscaled frames, and render final videos to disk.
    MIT