Every demo runs against the real protocol — no sandboxes. The agent demos need the wallet set up once; the rest need nothing but a browser.
https://wallet.browserid.me/mcp
Proves the core loop: your agent gets an identity, you approve it, and it signs a public wall with a line attributed to it and to you.
Provision a browserid identity and sign the guestbook with a fun message of your own.
Proves it works on a platform you don't control: an account of its own — or scoped posting to yours — with a badge on every post naming who authorized it and which agent wrote it. Sign in there and it hands you the prompt for your agent.
The kill switch, live: add the demo server as a connector (claude.ai: Settings → Connectors → Add custom connector), approve once, and ask your agent to log an action — attributed to it and to you. Then revoke at your account: the very next call is refused, fail-closed.
https://mcp-demo.browserid.me/mcp
Everything above proves the protocol — gate puts it to work. It publishes the MCP servers on your machine — notes, home automation, a database — at real URLs: your agent reaches them from anywhere, and you can share them by email with per-tool permissions, revocable anytime.
npx @browserid-ng/gate --admin you@example.com
A complete relying party in one file, and an MCP server whose tools require an agent identity + human-signed warrant.
Working prototypes probing where the protocol can go next. Everything here runs for real; the edges are rougher.
Experiment: can the browser's own account chooser (FedCM) improve vanilla browserid sign-in in Chrome transparently — same API for the site, no redirect, no popup?
Experiment: can browserid identities sign on-chain transactions and prove who's who on-chain? mingo.place is a tiny social app built that way — sign in with the email you already use (or your Bluesky handle, as me@your-handle), no password, and post.