andrew-babo/muse4ai

What it does

Muse4AI runs on a Windows PC and exposes a locally signed-in Muse session as an API that AI agents can call. Agents such as ChatGPT, Claude, Cursor, Codex, n8n, or custom scripts can request image and video generation, continue or extend clips, edit video through natural-language instructions, and maintain multi-turn chat with attached files. The app keeps the library, job queue, API keys, OAuth tokens and history on the machine under %APPDATA%\MuseStudio; requests go directly from that machine to Muse rather than through a relay server operated by the project README lines 30–50.

Agents connect in two ways. API keys are created in the app, scoped, rate-limited and revocable, and are sent as bearer tokens. OAuth support includes authorization-server metadata, dynamic client registration and a 6-digit consent code shown in the app, so a link alone cannot grant access OAuth setup. The REST surface covers job creation, listing, inspection and cancellation, asset and conversation extension, status and health; equivalent MCP tools are provided under names such as muse_generate, muse_chat and muse_extend_video agent tool mapping.

Quick start

You need Windows 10 or 11 and a Muse account. For the packaged path, download the installer from the download page, run it, and add the Muse account in Accounts. Windows reports the installer as unsigned; the README instructs choosing More info → Run anyway install steps.

To run from source, install Node.js 22+ and run from the repository root:

cd desktop
npm install
npm run build
npm start

The desktop app opens and serves its API at http://127.0.0.1:47821 source build steps.

Then open Settings → API & agent and either paste the generated key into the agent configuration, or add the MCP endpoint for OAuth:

{
  "mcpServers": {
    "muse4ai": { "url": "http://127.0.0.1:47821/mcp" }
  }
}

Approve access with the 6-digit code displayed in the app. For a direct REST call with an API key:

curl -X POST http://127.0.0.1:47821/v1/jobs \
  -H "Authorization: Bearer msk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"kind":"video","prompt":"a coffee cup on an oak desk, warm morning light"}'

The job enters the in-app queue and its result appears in the local library API key example. The machine-readable contract is available at /v1/openapi.json.

Constraints

• Unofficial automation. Muse4AI is not affiliated with Meta, and it automates a web session. Site changes can break it, and the README warns that automation may not fit Meta's terms of service honest notes.
• Windows-only packaging. macOS and Linux builds are not packaged or tested, though the core is described as platform-independent.
• Unsigned installer. The packaged installer triggers a Windows warning and is not signed; the source path avoids that distribution route.
• Local state. Results, keys and tokens are stored in %APPDATA%\MuseStudio on the same machine; there is no cloud account or telemetry server described.
• License. The project is MIT licensed license.

HotRepo's take

Muse4AI entered HotRepo's new-project list on 2026-11-04 with 198 stars and 54 forks reported over a growth window starting 2026-11-03, which means the repository was being observed in its first day and has no longer-term trend to assess. The technical pitch is a local bridge rather than a hosted service: the MCP endpoint, scoped keys and OAuth consent flow are concrete integration points, and both installer and source paths are documented. The main thing to watch is distribution trust, since the primary packaged route is an unsigned Windows installer whose relationship to the repository build is not described.

Development

From desktop/, npm test runs the test suite, npm run typecheck performs strict TypeScript checking, and npm run package:win builds Windows ZIP and installer inputs development commands.