Hushdrop

Quick start

The core way to use Hushdrop is from your agent — it publishes what it builds to your own domain.

Claude Code (recommended — auto-updates via the plugin marketplace):

/plugin marketplace add maxtechera/hushdrop/plugin install hushdrop

Cursor · Codex · Copilot · Gemini · 20+ Agent Skills hosts:

npx skills add maxtechera/hushdrop-skill

No install — open hushdrop.dev/try, drop an HTML file, get a link. No account.

One-off from the terminal — zero install, zero setup, no account:

npx hushdrop report.html       # → https://hushdrop.dev/u/xxxx (+ password, copied) — no setup, no account

Free hosted account — persistent links on your own handle (hushdrop.dev/you/<slug>):

npx hushdrop login                       # passwordless: GitHub or magic linknpx hushdrop report.html                 # → https://hushdrop.dev/you/report-a1b2 (persistent)

Your own domain — self-host on your Vercel Blob (free, MIT, your URL + brand forever):

# one-click: the "Deploy with Vercel" button above auto-provisions a Blob store — then:npx hushdrop-install                         # wire your agents + CLIhush deploy --domain share.yoursite.com   # or do it from a local clone

No dashboard required. Two commands to try; one more to own.

Why this exists

Anthropic shipped artifact sharing so teams could hand each other the things they make with AI. It's great — and it's locked to their surface, their domain, their account. Stacktree does the agent-native version, but it's a closed SaaS: your content sits on their servers, custom domains and limits are paywalled, and you can't audit or self-host it.

I make things all day with agents: reports, proposals, guides, dashboards, whole little sites. I wanted to hand someone a link that's mine — my domain, my brand, my keys — in the time it takes to type one command, without uploading client work to someone else's server. So hush does exactly that: terminal-native, zero-knowledge, on your own domain, usable from any agent. MIT.

Hushdrop vs. the alternatives

Full write-ups: vs Stacktree · vs Send · vs ShareDuo

Features

• 🔒 Zero-knowledge — branding + AES-256 happen client-side; the server only stores ciphertext.
• 🌐 Your domain, your brand — colors, logo, social cards flow into the unlock gate + previews.
• 🤖 Agent-native — one CLI command or 9 MCP tools; any agent that runs a shell can publish.
• ⏱️ Auto-expire & burn-after-read — --expire 7d, --burn, server-enforced cleanup.
• 📧 Email-gate & revocable share links — restrict by domain, mint/revoke guest links.
• 📦 Anything static — HTML, Markdown (rendered), PDFs, images, multi-file zips → sites.
• 🚀 One-click self-host — Deploy to Vercel auto-provisions Blob; no DB to run.

What you can drop

CLI reference

Use it from any AI agent

npx hushdrop-install registers the drops MCP server into your detected agents (Claude Code, Codex, Cursor, Windsurf, OpenCode, Amp), puts hush on your PATH, and prints config for GUI clients.

claude mcp add hushdrop -- npx -y hushdrop-mcp        # Claude Codecodex  mcp add hushdrop -- npx -y hushdrop-mcp         # Codex

{ "mcpServers": { "hushdrop": { "command": "npx", "args": ["-y", "hushdrop-mcp"] } } }

MCP tools (9 — vs Stacktree's 7)

How it works

1. Read your file and detect its type.
2. Brand (HTML) — inject favicon, OG/Twitter card, and a subtle corner badge before </body>.
3. Encrypt (if locking) — StatiCrypt (AES-256, client-side) behind your branded unlock gate. The badge is baked in before encryption, so it survives.
4. Upload to Vercel Blob under a clean key.
5. Serve — yourdomain.com/<slug> proxies the blob via middleware.js, rewriting headers so encrypted HTML decrypts + renders (not downloads) and CSP doesn't block the unlock script. Drops are noindex, nofollow, noai.
6. Report — URL (+ password) printed and copied to your clipboard.

Serving is a dumb transparent proxy; all branding + encryption happen client-side at upload. The server only ever stores ciphertext.

Self-host

[Deploy with Vercel]

The Deploy button clones the repo and auto-provisions a Blob store (it injects BLOB_READ_WRITE_TOKEN; the app derives its store from the token, so it just works). Then npx hushdrop-install + hush init --domain share.yoursite.com to point your CLI + agents at your instance. Self-host is token-only, free, unlimited — no account needed. Full walkthrough in skill/SETUP.md.

Configuration

FAQ

Is it really zero-knowledge? Yes — for locked drops, AES-256 runs in your browser before upload; the server stores only ciphertext. Managed vs. hosted vs. self-host? Managed = anonymous, 24h, no account. Hosted = free account, persistent links on hushdrop.dev/you/…. Self-host = your own domain + Blob. Does it need Vercel? Only to self-host. The managed + hosted tiers need nothing but npx. How is this different from Stacktree? Same agent-native idea, but open-source, self-hostable, your own domain free, and anonymous (no account). See the comparison. Can agents use it without MCP? Yes — it's a single CLI; any agent that runs a shell command can publish. Is it free? Yes. MIT. Self-host costs only your own (usually pennies) Vercel Blob usage.

Security

• Locked HTML is genuinely AES-256 encrypted in the browser — use long passwords; strong against casual access, not a vault.
• Raw files are protected by an unguessable slug; use --page -p <password> for a gated download.
• Passwords are stored in ~/.hushdrop/manifest.json on your machine only — never uploaded.

See SECURITY.md to report a vulnerability.

Contributing

PRs welcome — see CONTRIBUTING.md. It's a few hundred lines of readable Node + HTML, no build step.