Anthropic has built a direct counter to one of the more disruptive open-source tools in developer circles. Claude Code Channels, announced today, lets users connect Claude Code‘s agentic capabilities to Discord and Telegram, enabling asynchronous, on-the-go communication with an AI that can write and manage code independently.
The context matters. Since November 2025, an open-source project called OpenClaw — built by Austrian developer Peter Steinberger and initially named “Clawd” before Anthropic issued a cease-and-desist over trademark concerns — had been gaining rapid traction among developers and vibe coders. Its core appeal was simple: a persistent AI worker reachable 24/7 through everyday messaging apps including iMessage, Slack, Telegram, WhatsApp, and Discord, capable of performing substantive work autonomously — drafting and organizing email, building applications, applying for jobs on a user’s behalf, and managing social marketing campaigns — then notifying the user when a task was complete. Steinberger has since been hired by OpenAI.
OpenClaw‘s reach came with meaningful trade-offs. Granting an AI agent access to a user’s hard drive, file system, and personal data introduced security risks that non-technical users found difficult to manage, spurring a wave of derivative projects — NanoClaw, KiloClaw, and Nvidia‘s recently announced NemoClaw — each promising easier setup or tighter security controls.
What Anthropic is offering with Claude Code Channels is that same foundational capability — message the agent, let it work, receive an alert when it’s done — but packaged with the company’s established commitments to AI safety and a substantially lower barrier to entry for less technical users. Previously, Claude Code users were confined to interacting through the desktop application, terminal, supported developer environments, or a mobile interconnection setting called Remote Control, which the announcement describes as unreliable.
The architecture running beneath Claude Code Channels is the Model Context Protocol (MCP), the open-source standard Anthropic introduced in 2024 that provides a standardized method for AI models to connect with external data sources and tools. In the new Channels setup, an MCP server functions as a two-way bridge between the agent and external messaging platforms. When a developer starts a session using the --channels flag, the system spins up a polling service powered by the Bun runtime — selected for its speed executing JavaScript — that monitors the connected platforms. Incoming messages are injected into the active session as a <channel> event, prompting Claude to execute code, run tests, or fix bugs before replying via a dedicated reply tool. Critically, because sessions can run in a background terminal or on a persistent server such as a VPS, they do not time out the way standard web-chat interfaces do.
Setup requires Claude Code v2.1.80 or later and the Bun runtime installed on a desktop PC or Mac. For Telegram, users create a bot through BotFather using the /newbot command to obtain an access token, then install the plugin inside the Claude Code terminal. Official documentation covers the full configuration process for both platforms, according to the announcement.
Photo by Pixabay
This article is a curated summary based on third-party sources. Source: Read the original article
