wtf is broll?

Broll gives AI coding agents like Claude Code or Codex the ability to actually produce and publish social media content, not just write a plan for it. It connects to those agents through the Model Context Protocol, a standard that lets an AI assistant call out to extra tools. With broll, an agent can generate images and video, cut them into finished videos or carousel slides, and prepare a social media post, with a human confirming before anything actually goes live. Image and video generation uses your own OpenAI or Gemini API key directly, so broll never resells or marks up that usage, and there is a built in mock provider so you can try the whole flow without any keys at all. Rendering itself is deterministic: a video or carousel is described as a plan, and that plan is turned into exact ffmpeg or image rendering commands, so the same plan and inputs always produce the same output. That means things like fonts, captions, and branding live in code rather than being left up to unpredictable AI output. Publishing follows a draft first approach. Posts are written as reviewable files in a local outbox, and actually sending one out requires an explicit confirm setting, with checks for each platform's rules on character limits, media counts, and file sizes before anything leaves your machine. There is also an export option that always works, producing a ready to post bundle without needing any platform connection at all. Supported live platforms include Bluesky, Mastodon, and X, with LinkedIn and others planned. It is written in TypeScript, requires Node 20 or newer plus ffmpeg installed locally, and is installed by cloning the repository and building it with npm. The project is early but includes 96 tests, including real rendering checks. It is released under the MIT license, and a paid hosted version called broll cloud is planned for platforms that need extra approval steps.