Claude Channel CLI

claude-channel-cli

claude-channel-cli asks a live Claude Code session questions from Codex, Claude Code, or any local shell and returns Claude's response to the caller. Because it's built on the Channels feature, Claude Code can tell the request came from another agent and can respond through the channel without mixing that response into your normal chat.

Unlike using the API directly or a sub-agent, this gives you full auditability and steerability. For example, Codex can submit a plan to Claude for review while you keep watching and steering that same Claude Code session.

Install the CLI

Install the CLI from npm:

This makes the claude-channel command available on your PATH.

Requirements

- Claude Code 2.1.80 or newer.
- Claude Code authenticated with claude.ai or a Console API key.
- Claude Code Channels enabled. Team and Enterprise organizations must enable channelsEnabled.
- Channels are not available through Bedrock, Vertex, or Foundry providers.

Start Claude Code

Optionally check each project directory where you want Claude Code to receive channel requests. This is a read-only readiness and stale-registration check:

Then start Claude Code from that project with:

Pass normal Claude Code flags after start:

Use -- before Claude flags that would otherwise be handled by claude-channel, such as --version:

claude-channel start passes a session-scoped MCP config to Claude Code, so ordinary claude launches are not changed and do not start the channel receiver.

The receiver is intentionally not a standalone command. Start it through claude-channel start so the session-scoped MCP config and launch marker are generated together.

Upgrading from 0.3.x

If you previously used claude-channel setup-mcp or an early claude-channel setup, those versions wrote a…