Bridging Mobile and Code
Anthropic has launched an exciting new capability that allows developers to interact with their running Claude Code sessions directly from their smartphones,
using familiar chat applications like Telegram and Discord. This feature, dubbed Claude Code Channels, establishes a two-way communication link between a developer and their local Claude Code environment. Currently in a research preview phase, it supports Telegram, Discord, and a platform called Fakechat. To utilize this functionality, users must be running Claude Code version 2.1.80 or later and possess a claude.ai login. Notably, individuals who access the AI coding assistant through organizational API keys will not be able to leverage this new feature, indicating a focus on individual developer accessibility and a distinct approach to security and integration.
The AI Agent Landscape
The emergence of Claude Code Channels arrives at a time when the automation of software development is rapidly advancing, largely propelled by the growing popularity of agentic AI platforms. This trend is exemplified by systems like OpenClaw, which initially enabled users to run Claude Code through platforms such as WhatsApp on their mobile devices. However, the evolution of OpenClaw projects has expanded significantly, moving beyond simple mobile interactions to orchestrating complex AI agents capable of tasks like booking flights, managing smart home devices, and running social media campaigns across diverse platforms and devices in a structured manner. This broader application, while powerful, also brings into focus the inherent security considerations.
Navigating Security Concerns
The security implications associated with platforms like OpenClaw have spurred the development of safety-focused alternatives and forks within the AI agent ecosystem. Demonstrating this trend, Nvidia recently unveiled NemoClaw at GTC 2026, a software toolkit designed for integration with OpenClaw. Its purpose is to facilitate the secure execution of specialized AI agents, or 'claws,' within enterprise environments by confining them to controlled virtual settings. In contrast to Nvidia's approach, Anthropic's Claude Code Channels presents its own contender in the OpenClaw space, characterized by more stringent security measures and a more defined scope of operation. This launch also underscores Anthropic's competitive stance, particularly given that Peter Steinberger, the originator of OpenClaw, joined Anthropic's primary competitor, OpenAI, less than two months prior. Anthropic had previously issued a cease-and-desist letter concerning the initial naming of Steinberger's project, 'Clawd.'
Mechanism of Channels
Claude Code Channels operates on a novel principle, essentially inverting the typical Model Context Protocol (MCP) flows. In standard MCP, a user sends a request to Claude, which then determines the need to invoke a tool and forwards a request to the MCP server. The tool executes, returning data that Claude uses to formulate its final response. Channels reverses this process. A plugin for a messaging app, such as Telegram, connects to the platform's Bot API and actively monitors for incoming messages. Each 'Channel' effectively functions as an MCP server running locally alongside Claude Code as a plugin-backed process. When a new message arrives, it's encapsulated as a specific event and directed to the active Claude Code session. Claude Code then interprets the message, performs the necessary actions, and utilizes a reply tool to send the generated response back to the originating messaging platform like Telegram or Discord. The developer's terminal session displays the incoming message and a confirmation of processing, but the complete response is delivered directly to the external chat application. A key advantage is that Claude Code maintains the session's state across these interactions, eliminating the need to restart context for each new communication.
Practical Applications and Limits
Initial explorations with Claude Code Channels have showcased its utility for various tasks, including the development of iOS applications, the execution of command-line interface (CLI) tools, and the processing of audio files. These operations have been effectively managed from a smartphone, utilizing a platform like Telegram to remotely control Claude Code running on a host device, such as a Mac, without direct physical access. Despite these impressive capabilities, certain limitations have been identified. For example, when Claude Code requires user authorization for file operations or shell commands, the session temporarily halts. This necessitates the user to physically access their host device to manually approve the action within the terminal. While remote approval workarounds exist, they introduce potential security vulnerabilities, underscoring the trade-offs involved in enhancing convenience.
Subscription Access Details
Access to the Claude Code Channels feature is tiered based on subscription levels. For users subscribed to Claude Team and Enterprise plans, the Channels functionality is deactivated by default, requiring explicit activation. Conversely, individual subscribers of Pro and Max plans gain immediate access upon the feature's release. However, even for these users, there's a requirement to opt-in for Channels to be active at the commencement of each new session. This staggered rollout and opt-in mechanism highlight Anthropic's cautious approach to deploying new features and managing user permissions.
