Claude Code

Prerequisites: the riby and riby-mcp binaries installed and riby login completed on this machine. If not, start with the quickstart.

The MCP server needs no arguments and no environment variables. It reads its credentials from ~/.riby/config.json, which riby login writes.

Option 1: the CLI

One command, run anywhere:

claude mcp add riby -- riby-mcp

By default this registers the server for the current project. To make it available in every project on the machine:

claude mcp add --scope user riby -- riby-mcp

Option 2: .mcp.json

To check the server into a repo so every collaborator gets it, add it to the project's .mcp.json:

{
  "mcpServers": {
    "riby": {
      "command": "riby-mcp"
    }
  }
}

Each collaborator still runs riby login once on their own machine. Credentials never live in the repo.

Naming the session

Each running instance labels its pages folder · branch by default. To override, set RIBY_SESSION_NAME (100 characters max) in the server's environment:

{
  "mcpServers": {
    "riby": {
      "command": "riby-mcp",
      "env": {
        "RIBY_SESSION_NAME": "release-runner"
      }
    }
  }
}

Verify

Restart Claude Code (or start a new session), then run /mcp. The riby server should list four tools: request_permission, request_plan_check, request_steering, and check_pending. Ask for a test page:

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

Your phone buzzes; the tool call blocks until you answer. That block is the point — see why it holds.

Teach Claude when to page

Wiring up the server gives Claude the ability to page. The page-your-human skill gives it the judgment: when a decision warrants a page, when it does not, and what to do when a request expires. Install both.