Claude Code

Setup

1. Install the CLI

Any path from Install works — the install script, npm, or a GitHub release binary.

2. Register the MCP server

The zero-build path — chime mcp install writes the config for you:

chime mcp install            # writes ./.mcp.json (project-scoped)
chime mcp install --global   # writes ~/.claude.json (user-scoped)

Or register it directly with Claude Code’s own CLI:

claude mcp add chime -- chime mcp serve

Both produce the same mcpServers entry:

{
  "mcpServers": {
    "chime": {
      "command": "chime",
      "args": ["mcp", "serve"]
    }
  }
}

3. Pair your phone

chime pair

See Quickstart for the full pairing flow, including the codename match that’s your MITM check.

The plain-language path

Once Chime is installed and paired, you don’t need to invoke any tool by name. Claude Code auto-loads Chime’s agent runbook (SKILL.md), so you can just say what you want in plain language:

Send me a Chime when this finishes.

Set up chime.sh and notify me when you need my input.

Claude Code picks that up, runs the setup flow itself if it hasn’t already, and calls the right MCP tools without you naming them.

Why Claude Code is different

Claude Code is the only harness with a Monitor tool — the ability to supervise a long-running, never-exiting command (chime receive --watch) and surface each new line as a live turn the moment it arrives. That’s what makes replies feel pushed instead of polled. Every other harness in Integrations uses the portable chime_wait / chime_receive MCP tools instead — see MCP for how those work; it’s the same underlying loop, just without the push.

MCP — the six tools this integration is built on.
Quickstart — install, pair, first message, start to finish.