Why I Built Hushdrop: A Share Button for Your AI Agents

I make things all day with agents: reports, proposals, guides, dashboards, whole little sites. And every single time, the same friction showed up at the end — how do I hand someone the thing I just made?

Copy the HTML into an email? Spin up a Vercel project for a one-off page? Paste it into a tool that keeps my client's work on their servers, behind their login, on their domain? None of that felt right. 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.

So I built Hushdrop.

What it is

Hushdrop turns any artifact — an HTML page, a report, a Markdown doc, a file, a whole static site — into a private, branded, password-protected link on your own domain, in about a second. It works from your terminal, and more importantly, from inside your agent.

npx hushdrop report.html --managed
# → https://hushdrop.dev/u/xh6sm9gpgq  (+ password, copied)

The interesting part isn't the CLI. It's that your agent can do this itself. Hushdrop ships as an MCP server with nine tools (publish_html, update_site, set_password, set_expiry, set_email_gate, and more), so Claude, Cursor, Codex, Gemini — anything that speaks MCP or Agent Skills — can publish what it just built and hand you back a link without you touching anything.

Secure by default, yours when you want it

Two things I refused to compromise on:

Zero-knowledge. When a link is locked, the branding and AES-256 encryption happen in your browser, before upload (via StatiCrypt). The server only ever stores ciphertext — it never sees your content or your password. I tested this end to end: publish a file, fetch the served page, and the plaintext simply isn't there.

Ownership. There's a zero-setup managed tier (anonymous, auto-expires in 24h) and a free hosted account — but the real upgrade path is self-hosting on your own domain and your own Vercel Blob. It's MIT-licensed. Fork it, audit it, run it on your infra. No SaaS lock-in, no per-seat pricing, no paywalled custom domains.

It also does the stateful stuff you actually need: burn-after-read (the link self-destructs on first view — server-enforced, I checked: first request 200, second request 410), revocable share links, email-domain gates, and expiry.

One click for plain Claude.ai and ChatGPT

Not everyone lives in a terminal. So Hushdrop runs a remote MCP endpoint — https://hushdrop.dev/mcp — that you can add as a custom connector in Claude.ai or ChatGPT in one paste. After that, just ask Claude to "publish this as a link" and it does.

Install it into your agent

# Claude Code (auto-updates via the plugin marketplace)
/plugin marketplace add maxtechera/hushdrop
/plugin install hushdrop
 
# Cursor · Codex · Gemini · 20+ Agent Skills hosts
npx skills add maxtechera/hushdrop
 
# or one file, straight from the terminal
npx hushdrop report.html

Why open source

There's a whole category forming around "let agents publish what they make." Some of it is good, closed SaaS. But the thing I'm handing someone is often client work — and I don't want it sitting on a vendor's servers behind a subscription I have to keep paying to keep the links alive. The honest answer is the boring one: your infra, your keys, your code. Hushdrop is the open-source version of that idea.

It's live at hushdrop.dev, the source is on GitHub, and the whole thing — CLI, MCP, the remote connector, self-host backend — is MIT.

Drop something. Share it with who you want. On your own terms.