Quickstart

Four steps: install, log in, connect your harness, link a channel. A few minutes end to end.

1. Install the CLI and MCP server

Riby ships two binaries: riby (the CLI) and riby-mcp (the local MCP server your agent talks to).

npm install -g @riby/cli @riby/mcp

[PLACEHOLDER — final install command lands with the distribution decision; the binary names above are final.]

2. Log in

riby login

The CLI prints an activation link and a short code:

Open https://runitby.you/activate?code=XXXXXXXX
Enter code: XXXXXXXX
Waiting for approval

Open the link, sign in if you are not already, and approve the device. The CLI confirms with Logged in as <agent-name>. One login covers the machine; name it something recognizable with riby login --agent-name my-laptop if you like.

Check it worked:

riby status

You should see the server, your agent name, and a pending-request count of zero.

3. Connect your harness

Riby is a stdio MCP server, so any harness that runs MCP servers can page you. For Claude Code:

claude mcp add riby -- riby-mcp

For other harnesses, add riby-mcp as a stdio server in the MCP config. No arguments, no environment variables required. Full walkthroughs, including config file variants: Claude Code, Cursor, and headless/CI.

Each running instance identifies itself as folder · branch by default. Set RIBY_SESSION_NAME in the environment to override the label (100 characters max).

4. Link Telegram

1. Open runitby.you and go to Channels.
2. Tap Link Telegram. A one-time code opens a chat with the Riby bot and links it automatically.

Discord links from the same page with a slash command instead of a deep link. The dashboard itself is always available as a channel, no linking needed. Details, including unlinking and how answers sync: the channels guide.

5. Get your first page

Ask your agent to run something it should check with you first, or tell it directly:

Use the riby request_permission tool to ask me before deleting anything.

Your phone buzzes. Tap Approve. The agent continues, and the message updates in place with how and when it was resolved.

That is the whole loop. To make the agent page on its own judgment instead of on request, install the page-your-human skill. For what each tool accepts, see the MCP tool reference. If a page never arrives, see troubleshooting.