Anthropic's new step in the AI agent race
Anthropic has just introduced a new feature called Claude Code Channels, which allows developers to communicate directly with Claude Code sessions running on computers through messaging applications such as Telegram and Discord.
This feature marks a remarkable step forward in the trend of automating programming using artificial intelligence, as AI tools are increasingly deeply involved in the software development process.
How Claude Code Channels works
Claude Code Channels operates based on plugins (small software installed into the application) using MCP (model context protocol), creating a two-way communication channel between the user and the local Claude Code session.
Unlike the traditional mechanism where users send requests and AI responds, Channels reverses this flow. Messages from applications like Telegram will be received by the plugin, packaged into events and sent directly to Claude Code. From there, AI processes requests and responds back through the messaging platform.
Each channel is actually an MCP server running in parallel with the Claude Code session. As a result, the system can maintain a continuous working state, without having to restart every time there is a new request.
Remote programming control by phone
One of the notable points of Claude Code Channels is its remote control capability. Initial tests show that users can build iOS applications, run command lines or process audio with just a phone, while Claude Code works on servers like MacBooks.
This opens up a more flexible programming prospect, where developers do not need to directly sit in front of a computer to work.
Competition with OpenClaw
The launch of Claude Code Channels took place in the context of platforms like OpenClaw becoming increasingly popular. Initially, OpenClaw allowed Claude Code to run via WhatsApp, but later expanded into an AI agent ecosystem that could book air tickets, control smart homes or manage social media campaigns.
However, security concerns have fueled the emergence of safer alternatives.
At the GTC 2026 event (held from March 16 to 19 in San Jose, California, USA), Nvidia introduced NemoClaw, a toolkit to help operate AI agents in a closed business environment.
Compared to competitors, Anthropic chooses a more cautious direction, focusing on security control and a narrow operating range. This move also shows the ambition to compete directly in the field of AI agents.
Limitations and conditions of use
Currently, Claude Code Channels has just been released as a research preview, with limited support on Telegram, Discord and Fakechat.
The feature requires a version of Claude Code 2.1980 or higher and login through the account claude.ai.
A notable limitation is that when performing sensitive operations such as accessing files or running system commands, Claude Code will pause to ask users to confirm directly on the host device. This helps enhance security but also reduces seamlessness.
Team and Enterprise package users are currently not allowed to activate this feature by default, while individual Pro and Max package users can access it immediately but still have to choose to participate each time they use it.
