CzzzzzzJ/mdldm-maitian
Productivity & UtilitiesA macOS workbench that plans cross-platform video releases, fills upload forms via Chrome, and tracks views, followers and conversions.
What it does
Maitian (麦田) is a self-hosted publishing hub for creators who post the same video to Bilibili, Douyin, Xiaohongshu, WeChat Channels, YouTube, X and Instagram. It reads a release project (finished video, cover, per-platform copy and schedule), then drives a dedicated Chrome instance to open each platform's upload page, upload files and fill in the copy. It also records post links, collects per-episode views, engagement and follower counts, and can reconcile those numbers with registrations and revenue from your own site.
The design is deliberately conservative for Chinese platforms: Bilibili, Douyin, Xiaohongshu and WeChat Channels are only filled in, and the human clicks publish. X, YouTube, Instagram and your own site can be submitted automatically after the form is read back and verified. A deduplication check skips platforms that are already done.
The project is split across two repositories: this one runs the scheduler, database, web dashboard, Feishu notifications and data collection; platform adapters live in the companion Chrome extension mdldm-maitian-publisher, whose lib/ is loaded directly so adapters are maintained once. The publishing skill in skills/maitian-publish lets an AI assistant watch a video, write platform-specific copy and generate a validated publish.json.
Getting started
Prepare a Mac that stays on, Node.js ≥ 22.5 (the project uses node:sqlite) and Google Chrome. A NAS for release projects, a Feishu group bot and cloudflared are optional.
The documented one-line install downloads both repositories into ~/maitian/, wires up the publishing skill and starts a setup wizard:
bash -c "$(curl -fsSL https://raw.githubusercontent.com/CzzzzzzJ/mdldm-maitian/main/scripts/install.sh)"
The wizard asks a few questions, generates config.json, installs launchd autostart and opens the platform login pages. Afterwards open the URL it prints, http://localhost:8787/?key=<web.key>, and the browser remembers the key.
The manual route gives a faster way to see the result before configuring anything:
git clone https://github.com/CzzzzzzJ/mdldm-maitian-publisher.git
git clone https://github.com/CzzzzzzJ/mdldm-maitian.git && cd mdldm-maitian
node src/demo.mjs
node src/init.mjs
node src/hub.mjs doctor
node src/demo.mjs starts a fake-data dashboard at http://localhost:8788/?key=demo, so the screenshots can be explored without real accounts. node src/init.mjs is the setup wizard, and doctor checks platform login state and upload pages without uploading or filling anything. The README describes three levels of adoption: read-only data first, then publishing, then optional conversion tracking tied to your own site.
Usage constraints
- macOS only; Node.js ≥ 22.5 and Google Chrome are required.
- The system runs on an always-on Mac and operates through a logged-in Chrome profile. Credentials are not stored, but
chrome-profile/,data/andconfig.jsonshould be kept out of git. - Bilibili, Douyin, Xiaohongshu and WeChat Channels always require the human to click the final publish button; automation only fills forms. Xiaohongshu is handled through conservative scrolling and clicking rather than direct page API calls.
- Each video should be uploaded once per platform; the deduplication logic exists to lower risk-control triggers. Videos need
faststart(moov atom at the file head) or Xiaohongshu and Douyin may fail to parse the video. - Conversion tracking requires your site to expose two read-only endpoints matching the fields documented in
src/site.mjs; without a configured site, those cards and the conversion page are disabled. - Remote access, Feishu multi-dimensional table sync, the iOS Scriptable widget and the X long-article-plus-quote flow are optional additions with their own configuration steps.
- Licensed under MIT.
HotRepo's take
On the HotRepo new-project board for 2026-11-16, Maitian recorded 153 stars and 47 forks, gaining 47 stars and 13 forks in the sampled window starting at 13:16 UTC that day; the available star curve is just two samples about half an hour apart, so it shows a quick start rather than a trend. The project's real distinction is the split-by-risk publishing model: it automates only what the platforms tolerate and keeps the final click on the Chinese platforms, which matches the per-platform pitfalls documented in the README instead of promising full unattended posting. The author also links a paid AI Agent course site in the README, but the repository itself contains the scheduler, dashboard, data collection and skill, so the promotional section does not make it an ad page.
Written by AI from the project's README, updated . The project's own docs are the source of truth.