Anthropic has rolled out a significant upgrade to Claude Code, its powerful local AI coding agent, with the introduction of Channels — a research-preview feature that lets external messaging platforms push events and messages directly into an active, running Claude Code session.
Announced in late March 2026, Channels transforms Claude Code from a terminal-bound tool into something far more accessible and always-on. Users can now send instructions, tasks, or alerts from Telegram or Discord to a persistent Claude Code instance running on their Mac or PC — and receive replies back through the same channel — without needing to be physically at the keyboard.
This move directly addresses a long-standing gap: Claude Code has excelled as an agentic coding companion (writing, debugging, and iterating on code autonomously), but it lacked native transport to popular chat apps. Enthusiasts had previously built workarounds like cc-connect, but Anthropic's official implementation brings security, ease of use, and brand trust to the table — positioning it as a strong counter to similar open-source tools.
How Channels Work
Channels operate as MCP (Multi-Channel Protocol) server plugins that bridge external platforms to a live Claude Code session.
- Two-way communication — Claude receives messages as events and can reply directly through the channel (e.g., sending code snippets, status updates, or results back to Telegram/Discord).
- Session persistence required — Events only arrive while the session remains open. For always-on behavior, run Claude Code in the background, via tmux/screen, or on a dedicated machine.
- One bot per session — Each configured channel (Telegram bot or Discord bot) corresponds to one active session. Multiple channels can be enabled simultaneously.
- Minimal host visibility — On the machine running Claude Code, inbound messages appear as structured `<channel>` events in the terminal, and outbound replies show as tool calls + "sent" confirmations — but not the full reply text, which is delivered externally.
Supported platforms in the initial research preview include Telegram and Discord, plus a fakechat demo for local testing (accessible via http://localhost:8787).
Setup Is Straightforward but Requires Care
1. Install the plugin
`/plugin install telegram@claude-plugins-official` (or discord)
Update the marketplace if needed: `/plugin marketplace update claude-plugins-official`
2. Configure credentials
Create a bot (BotFather for Telegram, Developer Portal for Discord), obtain the token, then run `/telegram:configure <token>` (similar for Discord).
3. Launch with Channels enabled
`claude --channels plugin:telegram@claude-plugins-official`
(Multiple plugins can be space-separated.)
4. Pair your account securely
Send a DM to the bot → receive a pairing code → approve it in Claude Code with `/telegram:access pair <code>` → set an allowlist policy.
Only allowlisted sender IDs can interact; others are silently dropped.
Security & Permissions: Thoughtful but Not Frictionless
Anthropic has clearly prioritized safety:
- Sender allowlisting prevents unauthorized access.
- Pairing requires explicit approval inside the trusted Claude Code terminal.
- Enterprise/Team plans need admin enablement via claude.ai settings.
- Permission relay capability (if supported by the channel) forwards tool-use prompts back to the chat for remote approval.
However, a key limitation remains: if Claude needs user permission for a sensitive action (file write, command execution, etc.), you cannot grant it remotely via Telegram or Discord.
You must either:
- Be physically at the terminal, or
- Start the session with `--dangerously-skip-permissions` (only recommended in fully trusted environments).
This design choice prevents escalation risks but reduces "true headless" usability for complex workflows.
Also read:
- OpenAI Is Building a Desktop “Super App”: Merging ChatGPT, Codex, and Atlas into One Unified Platform
- ElevenLabs Launches Music Marketplace: Creators Can Now Sell, Remix, and License AI-Generated Tracks
- Criminal Charges for AI-Generated Music? The First U.S. Case Sends a Clear Warning
- Google Just Supercharged Stitch — And Turned It Into the Ultimate “Vibe-Design” Tool
Why This Matters — And Why Now
Channels is a clear response to community demand for agentic AI that lives in everyday chat apps. Tools like OpenClaw (and earlier hacks) already offered similar remote control, but Anthropic's version arrives with official plugins, audited code (from claude-plugins-official GitHub repo), and Claude's strong reasoning & coding capabilities.
Early users report success with scenarios like:
- Remote podcast transcription;
- On-the-go code reviews;
- Receiving CI pipeline results or alerts directly in chat;
- Quick prototyping while away from the desk.
With nearly 14 million songs already generated using the music model, the team is confident the Music Marketplace will replicate — and potentially exceed — that momentum.
As a research preview (requiring Claude Code v2.1.80+ and rolling rollout), expect syntax and protocol changes. But the foundation is solid: Claude Code is evolving from a powerful local agent into a truly mobile-accessible, always-available coding collaborator.
Time to fire up a persistent session and test it — just be cautious with permissions. The future of agentic coding just got a lot more conversational.
