Explained: How Anthropic’s Claude Code Channels Let Developers Control AI Coding Sessions From Their Phones

Hyderabad: Dario Amodei-owned Artificial Intelligence (AI) company, Anthropic, has introduced Claude Code Channels for its AI chatbot, Claude AI. The new feature enables developers to send messages to an active Claude Code session, which is directly running on their computer, from messaging apps such as Telegram and Discord on their phones.

The Claude Code Channels is currently available as a research preview tool and supports Telegram, Discord, and Fakechat. It requires Claude Code v2.1.80 or later, along with a claude.ai login to function. Developers who access Claude Code through organisational Application Programming Interface (API) keys are not eligible to use it.

How Claude Code Channels work

The Claude Code Channels operate by inverting the standard Model Context Protocol (MCP) flow — an open-source standard created by Anthropic to allow AI models (LLMs) to seamlessly connect with external data sources, tools, and systems. In a typical MCP setup, a user sends a request to Claude, which calls a tool, queries the MCP server, and returns a response. Channels reverse this process.

Flowchart showing how Claude Code Channels work (Image Credit: Channels Reference/Claude Code Docs)

Each Channel functions as an MCP server running locally alongside Claude Code as a plugin-backed process. When a message arrives on Telegram or Discord, it is wrapped as a Channel event and routed to the active Claude Code session.

Claude Code reads the message, completes the requested task, and sends the reply back to the messaging platform. It is worth noting that Claude maintains session state across events, meaning it does not reset each time a new terminal window is opened.

Early use cases have included building iOS apps, running Command Line-Interface (CLI) tools, and processing audio — all managed remotely from a phone via Telegram.

Claude Code Channels: Limitations and Access

The Clause Code Channels feature also includes several limitations. The feature will pause a session whenever it requires user approval for a file operation or shell command, which currently requires the developer to physically interact with the host device. Remote approval workarounds exist but carry security trade-offs.

Flowchart on how Claude Code Channels request permissions from human to work. (Image Credit: Channels Reference/Claude Code Docs)

The Claude Code Channels are disabled by default for Team and Enterprise subscribers. Pro and Max individual subscribers receive immediate access, though they must still opt in at the start of each session.

Claude Code Channels - A competitor to OpenClaw

The launch of the Claude Code Channels comes amid growing developer interest in agentic AI platforms. The new feature bears some resemblance to early OpenClaw setups, which enabled users to run Claude Code through WhatsApp. OpenClaw has since expanded into orchestrating AI agents for tasks including flight booking, smart home control, and social media management.

Security concerns around OpenClaw have prompted the development of safety-focused alternatives. Last week, at GTC 2026, Nvidia introduced NemoClaw, a software toolkit designed to help specialised agents operate safely within enterprise environments via a contained virtual environment.