flow-mcp
Drives Google Flow in a local Chrome window to make videos, images, narration and captions
Chat with Meta's Muse agent through your own logged-in Chrome session
By duclm1x1 (individual) JavaScript AI and MLBrowser automation
23 popularity score ★ 76
Muse-Chat-MCP lets an MCP client send prompts to Meta Muse, the muse.ai "Hatch" agent, and read its replies. Muse has no public API and its chat runs over an encrypted WebSocket, so this project drives the user's real, logged-in Chrome with playwright-core: it types into the Muse composer and reads the rendered reply.
Over stdio it offers tools to check login and browser state, wait for sign-in, start a new chat, send a prompt (with optional image, video or document attachments), list, open and read chats, pull out links to images and videos Muse generates (optionally downloading them), dump the page DOM for selector debugging, and close the browser. The same process also starts an OpenAI-compatible HTTP shim on 127.0.0.1:8787 (/v1/models, /v1/chat/completions with streaming and prompt-injected tool calling), and a small CLI talks to that shim.
It needs Node.js 18 or later, an installed Google Chrome or Edge, and a Meta account with access to Muse. It is installed by cloning the repository and running npm install; the first run opens Chrome so you can sign in, and the session is kept in a dedicated profile folder. Requests run one at a time, and a change to the Muse web interface can break it. It is an unofficial project, not affiliated with Meta.
muse_statusmuse_loginmuse_new_chatmuse_chatmuse_read_lastmuse_chatsmuse_open_chatmuse_read_chatmuse_mediamuse_dump_dommuse_close