mcp server

ittybittyvideos

tiny tool, big videos. topic in, video out. no film degree required. webapp, windows app and mcp server for agent use

sandraschi/ittybittyvideoscommunitylistedunreviewed

Use tool

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

uv (`curl -LsSf https://astral.sh/uv/install.sh | sh`)

~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "ittybittyvideos": {
      "command": "uvx",
      "args": [
        "videogen-mcp"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 21, 2026

ittybitty

Topic or script in → narrated video out. Shorts for TikTok and Reels, or longer chaptered explainers — with stock footage, your Jellyfin/Plex home videos, or local AI clips.

· 中文 README

Beta. Core flows (generate, stock, MCP, Windows installer) are usable; R3 screening, R9 talker, and APIs may still change between releases. Report issues on GitHub.

Three ways to use it (pick one):

	Windows app	Web dashboard	MCP server
For	Everyday use on your PC	Browser UI while you develop or self-host	Cursor, Claude, other MCP clients
You get	Installer + desktop shortcut; backend starts automatically	Generate, Jobs, Depot, Settings, Help	`videogen_*` tools for agents
Start	Download installer	Run locally or use the app above	Connect MCP

Same engine under all three: Python backend on port 11054, React dashboard on 11055 when running the dev stack.


Windows app	Installer — no Python required
Web dashboard	Browser UI (dev or bundled in desktop app)
MCP server	Agent automation
What you need	Keys and optional extras
Footage sources	Where B-roll comes from
vs MoneyPrinterTurbo	Why pick ittybitty
More help	Config, troubleshooting, API docs
Marketing site	GitHub Pages one-pager (also website/ locally)

Windows app

Best if you just want to make videos.

Download the latest installer from Releases (tag v0.2.x or newer)
Run the installer → launch ittybitty from Start or your desktop shortcut
In Settings, add a free stock API key (Pexels, Pixabay, or Coverr) → Generate with a topic or paste a script

The installer bundles the web dashboard and Python backend — no separate Python or Node install.

Install folder: %LOCALAPPDATA%\ittybitty\ If the backend fails to start: %LOCALAPPDATA%\ai.fleet.ittybitty\logs\backend-spawn.log

Web dashboard

The React UI: Generate, Plan (mid-length), Jobs, Depot, Publish, Settings, Help, Logs.

Already using the Windows app? You're in the dashboard — nothing else to install.

Running on your dev machine (live reload, for contributors):

git clone https://github.com/sandraschi/ittybittyvideos.git videogen-mcp
cd videogen-mcp
.\start.bat

Open http://127.0.0.1:11055. The script starts the dashboard and the API together.

Needs Python 3.10+, FFmpeg, and Node.js — see INSTALL.md.

MCP server

For AI agents that should generate or plan videos over the Model Context Protocol.

Start the backend (Windows app running, or .\start.bat -BackendOnly, or uv run python -m videogen_mcp.server)
Point your MCP client at http://127.0.0.1:11054/mcp

Example prompts: "Generate a 45-second video about Vienna coffee culture" · "Plan a 5-minute sourdough tutorial" · "Call videogen_help first"

Tool list and REST API: docs/TOOLS.md (16 MCP tools; catalog at /api/v1/tools) · Claude Desktop / MCPB: INSTALL.md

What you need

For	You need
Most workflows	FFmpeg on PATH + free stock API key (Pexels, Pixabay, or Coverr)
Topic → script	DeepSeek or OpenAI key, or paste your own script
Home videos as B-roll	Jellyfin or Plex URL + token (Settings)
Local AI clips (GPU)	CUDA ~24 GB + `.\start-localgen.bat`

The in-app Help page walks through each step.

Footage sources

Pick one in Settings → Footage (free tier = no GPU):

Pexels — free stock (default)
Pixabay — free stock (MPT parity)
Mixkit — free 1080p clips, no API key
NASA — public-domain space/science footage, no API key
Jellyfin / Plex — cut clips from your library (vacation, pets, …)
Veo / Omni — Google cloud (config)
LocalGen — Wan 2.2 on your GPU

Finished videos land in ./output/ and appear in Depot.

How this differs from MoneyPrinterTurbo

MoneyPrinterTurbo popularized topic → short video. ittybitty targets the same job with fleet tooling: agents, longer formats, your own libraries, and a path to local GPU — beta quality: solid for daily use, still evolving.

	MoneyPrinterTurbo	ittybitty (beta)
How you run it	Web UI	Windows app + web dashboard + HTTP MCP
Length	Mostly shorts (~60s)	Shorts + 3–15 min chapters (planner + videographer rules)
Footage	Pexels, Pixabay, Coverr (+ local/social)	Same three + Mixkit + NASA (no key) + Jellyfin/Plex + LocalGen + Veo/Omni
Agents	—	*`videogen_` tools** for Cursor, Claude, fleet MCP clients
Library & jobs	—	SQLite depot, job history, publish handoff
Edit intelligence	Assembly / concat	Hook, pacing, B-roll, transitions; R1 karaoke subs; R2 beat snap + ducking; R3 screening (experimental)
Architecture	Monolith	Plugin registry — swap LLM, stock, and TTS providers
Desktop	—	NSIS installer (Windows, backend bundled)
Status	Mature OSS	Beta — CHANGELOG

中文说明与中国本地栈（通义千问、CosyVoice 等）：README-zh.md

Sample output

"German Shepherd puppy discovers snow for the first time" -- generated in ~55 seconds. Download full MP4 (17 MB)

Quick test render:

py scripts/smoke_render.py

docs/examples/README.md

More help

Doc	When to read it
INSTALL.md	All install paths, MCPB, verification
docs/CONFIGURATION.md	Env vars and providers
docs/TROUBLESHOOTING.md	Something broke
docs/PROMPT-DIRECTOR.md	R10 trope templates (mermaid)
docs/EXEMPLARS-RESEARCH.md	Viral Short formats + mid expansion
docs/TOOLS.md	MCP tools and REST API
docs/DEVELOPMENT.md	Tests, Tauri build, CI, pre-commit
SPEC.md	Architecture and roadmap
CHANGELOG.md	Version history

Fleet docs: mcp-central-docs/projects/ittybitty

MIT · sandraschi · v0.2.0

Repo folder videogen-mcp, Python package videogen_mcp — kept for MCP tool compatibility.