Lazycow Studio

lazycow

Lazycow Studio from the command line, from your AI agent and from your code.

Lazycow Studio turns the photos and videos you never edited into short vertical reels and YouTube trip films. It runs on your own computer, and a person approves every post. This package connects to it in three ways:

You want to…	Run
Install the studio (no terminal skills needed after this line)	npx lazycow install
Let Claude Code, Claude Desktop or Cursor drive it	claude mcp add lazycow -- npx -y lazycow mcp
Script it from Node	import { createClient } from "lazycow"

Approving a reel stays a person's job in the studio. Agents and scripts can make, plan and post approved reels. They cannot approve.

Watch the 40-second demo · See the studio's pages

Install the studio

npx lazycow install

It downloads this version's release zip from GitHub, checks it against the SHA-256 built into the package, unpacks it to ~/Lazycow Studio, and starts the double-click setup. The setup installs Node, Python and ffmpeg for Lazycow only, in about 10 minutes. Options: --dir <empty folder>, --yes (take every suggestion).

Prefer to click? Download the zip from Releases, unzip it, read README.TXT, and double-click Start Lazycow.

Connect an AI agent (MCP)

Claude Code:

claude mcp add lazycow -- npx -y lazycow mcp

Claude Desktop or Cursor (claude_desktop_config.json / .cursor/mcp.json):

{
  "mcpServers": {
    "lazycow": { "command": "npx", "args": ["-y", "lazycow", "mcp"] }
  }
}

The server finds the studio by itself: STUDIO_PORT and STUDIO_API_TOKEN when set, else the .env in LAZYCOW_DIR, else the folder the setup recorded. So your agent config never holds a copy of the token.

The 27 tools ("reads" change nothing; "spends" means paid AI, priced first and kept inside the studio's limits):

Area	Tools
The studio	studio_status, list_projects, switch_project, read_log (reads, except switching)
Reels	list_reels, reel_details, list_themes (reads); make_reels, make_reel, remix_reel (render in the background)
Jobs	job_status, cancel_job
Posting	list_posts, list_accounts (reads); plan_post, publish_reel, schedule_reel (approved reels only)
Research	research_ideas, research_events, research_page, research_site_check, research_links (reads the web through the studio; suggestions, never posted)
AI assist	ai_estimate, ai_review (reads); ai_run (spends, never changes an approved reel)
Results	perf_summary, perf_suggest (reads)

Every tool, its inputs and examples: docs/mcp.md.

Each tool answers with a plain sentence and the data, and the same data as structuredContent for programs (a list arrives as { "items": [...] }). Account secrets never leave the studio.

Listed in the official MCP Registry as io.github.kensaurus/lazycow.

Use the API from code

import { createClient } from "lazycow";

const studio = createClient();            // or createClient({ port: 4318, token: "..." })
const health = await studio.health();
const { reels } = await studio.overview();
const anything = await studio.request("GET", "/api/posts");

Every route is in the studio's OpenAPI document. Errors are StudioError with status, path and body. Types are included.

From a shell:

npx lazycow status
npx lazycow api GET /api/posts
npx lazycow open

Requirements

Node 20 or newer. The studio itself runs on Windows 10/11 and macOS 14+ (Apple silicon); Linux works from source.

Credits and licence

Made by kensaurus (Kenji Sakuramoto, kensaur.us). Apache-2.0: if you redistribute it or build on it, keep the NOTICE file, and please keep the line "Based on Lazycow Studio by kensaurus". "Lazycow" and the sleepy-cow icon are trademarks of kensaurus LLC (TRADEMARKS.md).

Like it? Support Lazycow: supporters get their feature requests looked at first.