What is Claude Code Channels, Anthropic’s take on OpenClaw-style AI agent setups?

Anthropic’s latest Claude Code feature comes amid the widespread automation of software development workflows, fueled by the surging popularity of agentic AI platforms such as OpenClaw. Similar to Channels, early setups of OpenClaw were focused on running Claude Code through WhatsApp on your phone. However, OpenClaw developer projects have since evolved into orchestrating AI agents to book flights, control smart home devices, and manage social media campaigns across multiple platforms and devices in a structured manner.

Yet, the security tradeoffs associated with OpenClaw have sparked the need for safety-focused forks and alternatives. At last week’s GTC 2026, Nvidia introduced a software tool kit called NemoClaw that is designed to be integrated with OpenClaw in order to help specialised agents or ‘claws’ run safely in an enterprise context, via a contained virtual environment.

Unlike the chipmaker, Anthropic has shipped its own OpenClaw rival with tighter security controls and narrower scope. The Channels launch also highlights its competitive posture, arriving less than two months after Peter Steinberger, the creator of OpenClaw, joined Anthropic arch-nemesis OpenAI. Anthropic previously sent a cease-and-desist letter to Steinberger over the original OpenClaw project name ‘Clawd’.

How does Channels work? What can it do?

Channels is reportedly based on an inversion of MCP (Model Context Protocol) flows, where a user sends a request to Claude, which decides to call a tool and send a request to the MCP server. The tool runs and returns data, which Claude turns into its final response to the user query.

Channels work the other way around. Telegram’s plugin is connected to the Bot API and scans for incoming messages. Each Channel is actually an MCP server running locally alongside Claude Code as a plugin-backed process. When a new message is sent, it gets wrapped as a gt; event and routed to the active Claude Code session. Claude Code reads the message, executes whatever work the message requires, and calls a reply tool that sends the response back to Telegram or Discord.

The open session on the terminal shows the inbound message and a confirmation, but the final reply only appears on the external platform. Claude keeps the state of the session the same across events, which means that it does not have to start from scratch every time someone opens a terminal window.

Early experiments with Claude Code Channels involve building iOS apps, running CLI tools, and processing audio – all worked through on a phone with Telegram installed effectively using it to run Claude Code on a host device such as a Mac wirelessly.

However, there are some limitations that have been pointed out as well. For instance, Claude Code pauses the session when it needs user approval for a file operation or shell command, requiring the user to walk to their host device and click on ‘Allow’ within the terminal. While there are workarounds that allow users to grant approval remotely, they come with security tradeoffs.

Claude Team and Enterprise subscribers have Channels disabled by default, but Pro and Max individual subscribers get access immediately. Though, they still have to opt-in every session.