Claude Code Channels Let Users Control AI Coding Agent Directly via Telegram

Claude Code’s new “Channels” feature turns a developer’s phone into a remote control panel for an AI‑powered coding assistant. By exposing a Model Context Protocol (MCP) server that can ingest messages from external services, Claude Code lets a running session listen for Telegram or Discord events, act on them, and push replies back through the same channel. In practice, a programmer can launch a long‑running refactor, step away from the terminal, and later send a “skip this file” or “add unit tests” command from a Telegram chat; Claude reads the instruction, updates the code, and replies with the diff—all without reopening the IDE (Alan West, Claude Code Channels).

The implementation relies on a lightweight plugin architecture that ships with Claude Code v2.1.80 and later. After creating a bot with Telegram’s @BotFather and installing the “te” plugin, the MCP server binds the bot token to a sender allowlist, ensuring only authorized users can inject events. The channel operates bidirectionally: every incoming message becomes an event in the Claude session’s context, and Claude’s response is emitted as a Telegram message, preserving the conversational flow. The documentation notes that the feature is “background‑friendly,” meaning the AI continues processing even when the user’s terminal is idle, and it is currently in a research preview, signaling that Anthropic is still gathering feedback before a broader rollout (Claude Code documentation).

Security and usability were front‑and‑center in the design. The allowlist mechanism prevents rogue bots from hijacking a coding session, while the requirement for a local Claude Code installation (rather than API‑key authentication) keeps the execution environment under the developer’s control. The setup steps are deliberately minimal: verify Claude Code’s version, install the Bun runtime, create a Telegram bot, and run a single plugin command. Once configured, the channel can be invoked with simple slash commands, letting developers push “push‑to‑origin” or “run tests” instructions from anywhere with an internet connection (West).

Anthropic’s broader roadmap underscores why Channels matter. The same week the feature landed, the company announced Claude Sonnet 4.5, a model billed as a “coding beast” by CNET, and Claude Opus 4.6, which expands the context window to one million tokens and introduces “agent teams” for more complex workflows (CNET; VentureBeat). By pairing a more capable model with a remote‑control interface, Claude Code positions itself as a full‑stack AI development partner that can handle everything from incremental edits to multi‑step project scaffolding without the developer ever needing to stare at a terminal. The synergy between larger context windows and real‑time, out‑of‑band messaging could reduce context‑switching friction, a perennial pain point for engineers juggling code, tickets, and chat.

Early adopters are already testing the limits of the workflow. One developer described using Channels to debug a microservice while commuting: they would send a “log current request payload” command via Telegram, receive the JSON dump in the chat, and then issue a “patch error handling” instruction—all before stepping off the train. While anecdotal, such use cases illustrate how the feature could reshape the ergonomics of AI‑assisted development, especially for remote or distributed teams that rely heavily on messaging platforms. As Claude Code continues to iterate—adding Discord support and refining the MCP server—the real test will be whether the convenience of phone‑based control outweighs the overhead of maintaining bot tokens and allowlists in production environments.

If the preview proves stable, Channels could become a template for other AI‑driven tools seeking tighter integration with everyday communication apps. The concept of “pushing events into a running model session” is not limited to coding; it could be repurposed for data analysis, content generation, or even customer‑support bots that need to act on live inputs. For now, Claude Code’s Channels give developers a tangible glimpse of a future where AI assistants are as reachable as a text message, blurring the line between local development environments and the always‑on chat ecosystems that dominate modern work.