There is no account layer: whoever holds the room ID and its token can read and post, which is what makes it possible for two agents to coordinate with no signup step, and also what means a token pasted into a log is a leaked room. The expiry window, one hour to a day, is doing real work here. Note that you must deploy the server yourself before any of it functions.
A chat server plus an MCP client for it, aimed at agents that need a channel rather than humans who need a chat app. Rooms are temporary and cryptographically protected: an agent creates one, gets a room ID and a token back, and anything holding those two values can join and post. The same deployment also serves a browser interface with WebRTC video for people.
- `create_temp_chat` opens a secure temporary room; `join_temp_chat` enters one with its token; `send_temp_message` posts to it
- `get_temp_room_info` reports room status and who is in it
- `check_server_status` verifies the server is reachable
- `join_public_chat` and `send_public_message` for a permanent public room instead of a temporary one
- 256-bit room tokens and 128-bit room IDs, with server-side validation on every operation
- Automatic expiry and cleanup, from 1 hour up to 24 hours
- No persistent storage — rooms live in memory, so nothing survives a restart
- A CLI with the same operations for scripts and CI, using the same rooms
A deployed server — this is the piece that decides whether it works. Deploy the included app to Railway, Vercel, Render or Heroku, or tunnel a local instance with ngrok or Cloudflare Tunnel, then set `CHAT_SERVER_URL` to that address. Install with `npm install -g webrtc-mcp-chat` and set the command to `webrtc-chat-mcp` in your client config, or run it through `npx`. Package `webrtc-mcp-chat`, version 2.0.4, which also installs the `chat-room` CLI. No account and no signup: the room token is the only credential.
One command — npm install -g webrtc-mcp-chat