Live reload — see your AI write in real time

What it does

MD View is a live reload markdown reader for Windows: open a file, and the app starts watching it on disk. Anything that writes to it — your editor, an AI coding agent, a CI job, a Notion export, a cat > foo.md from a shell — triggers a re-render in the open window. There’s no refresh button. There’s no toggle. The view is always the file as it currently exists. If you scrolled to line 200 before the reload, you’re still at line 200 after; if you were focused on a specific heading via anchor link, that anchor wins.

The watcher is debounced (250 ms) so a flurry of writes from a streaming AI doesn’t trigger a re-render storm — you see the file settle, then update.

Why it exists

If you’ve ever asked Claude (or any AI coding agent) to write a long document — a plan, a spec, a checklist — you know the alt-tab problem. The agent streams to a file in your repo. You want to read along. Your editor doesn’t auto-reload the rendered view. Your terminal shows raw markdown, not headings and bullets. So you alt-tab between the agent’s output and the rendered view, hitting refresh, breaking your concentration.

MD View just watches the file. You open it once. The agent writes. You read.

How to use it

There’s nothing to enable. Open any file with MD View — by double-click, file dialog, or mdview foo.md — and live reload is already on for that file. If the file is deleted while open, MD View shows a “File not found.” message in place of the document until you reopen something.

Under the hood

The watcher is a Mutex<HashMap<PathBuf, Debouncer>> in Rust, using the notify-debouncer-full crate. There’s only ever one debouncer per watched file path — opening the same file twice is idempotent. Multi-path support exists for split view (two open documents, two watchers).

On the JavaScript side, every watcher event is treated as “something changed → re-read the file.” There’s no attempt to interpret typed event semantics (created vs modified vs renamed) — Windows filesystem events lie about that constantly, especially when editors do atomic-rename-on-save. Re-reading the file is cheap and always correct.

Use cases

• AI agents writing long-form output: plan.md, spec.md, CLAUDE.md — open in MD View, let the agent stream.
• Build pipelines that produce reports: open coverage.md or bench-results.md, run the pipeline, read the report as it’s regenerated.
• Note-taking from another editor: keep your favorite editor in one window, MD View in another, edit in the editor, read in MD View. No reload step.
• Live-rendered changelogs: open CHANGELOG.md while you’re writing it; see the formatting as you type in your editor.