Connect an agent

Send one message

The setup console (and its Agent page) gives you a message like this:

Hi! Please connect yourself to my Molty.

Instructions: https://<relay>/skill.md
Device: molty-xxxxxxxxxx
Code: K7M2W9C4

I'll tap Molty to accept you.

Paste it into Claude, Grok, OpenClaw, or any agent that can make HTTPS requests. The skill.md link is the agent’s own guide: it explains what Molty is, how to pair, and every command.

Tap to accept

The agent asks the relay to pair. Molty condenses a 4-digit code out of the fluid. If your agent shows the same code, tap the screen to accept. Press PWR to reject. Nothing happens without that tap.

The agent receives a token scoped to this one Molty. An agent can revoke its own token; forgetting the link on the device (POST /api/link/forget) revokes every agent at once.

What your agent can do

Command	Effect
`mood`	idle, listening, thinking, speaking, happy, sleepy, surprised
`show`	speed-read text in the fluid
`say`	speak through Molty’s speaker while the words flow on screen
`listen`	record from the mic and return the transcript
`status`, `imu`, `volume`	read state, motion, set volume

say and listen return a job ID right away; the agent polls for the result.

Three ways in

Remote MCP — https://<relay>/mcp with the token as a bearer. Tools are named molty_pair, molty_say, molty_show, and so on.
REST — POST /v1/devices/<id>/cmd {name, args} and GET /v1/jobs/<job>.
CLI / local MCP — the molty binary, either through the relay or directly over USB-C / Wi-Fi with no cloud at all. For Claude Code:
Terminal window
```
claude mcp add molty -- /path/to/molty mcp
```

Full details: Agent link protocol and CLI and local MCP.

Run your own relay

The default relay is a hosted Cloudflare Worker. The relay is open source too; deploy your own and point the device at it from the console’s Agent page (Advanced). See the default relay section.