Claude Channels Scores 375 Points on Hacker News — Anthropic’s Play to Replace OpenClaw

On March 20, Anthropic shipped something that changes how developers interact with Claude Code. Claude Channels is a research preview feature that pushes real-time events — CI builds failing, monitoring alerts firing, Telegram messages arriving — directly into a running Claude Code session. Instead of waiting for you to type a command, Claude now listens, reacts, and responds on its own.

The reaction was immediate. Claude Channels hit 375 points and over 220 comments on Hacker News within hours, landing at the #2 spot on the front page. VentureBeat called it an “OpenClaw killer.” Developer BentoBoi declared: “Claude just killed OpenClaw with this update. You no longer need to buy a Mac Mini.” And AI YouTuber Matthew Berman put it more bluntly: “They’ve BUILT OpenClaw.”

Whether those claims hold up depends on what you need from an always-on AI agent. But the direction is clear: Anthropic wants Claude Code to stop being a tool you pick up and put down, and start being a collaborator that works while you sleep.

What Claude Channels Actually Does

A channel is an MCP (Model Context Protocol) server that pushes events into your running Claude Code session. This is a fundamental architectural distinction — channels are not tools that Claude calls. They are external systems that push information to Claude, and Claude decides how to act on it.

The flow works like this: you set up a channel (say, a Telegram bot), connect it to your Claude Code session, and walk away. When someone sends a message to your bot, or a webhook fires from your CI pipeline, that event lands in Claude’s session as a <channel source="..."> event. Claude reads it, takes action — maybe fixing a failing test, maybe replying to the message — and optionally sends a response back through the same channel.

Channels are two-way by default. Claude doesn’t just receive; it can reply through Telegram, Discord, or whatever channel delivered the event. In the terminal, you see the inbound message and a confirmation that Claude replied, but the actual reply text appears on the other platform.

Three channels ship with the research preview:

Telegram — Create a bot via BotFather, configure the token in Claude Code, and pair your account with a security code. Claude then responds to messages sent to your bot.
Discord — Similar setup through the Discord Developer Portal. Create a bot, enable Message Content Intent, invite it to your server, and pair.
Fakechat — A localhost demo that runs a browser-based chat UI. No external services, no authentication — just a quick way to see how channels work before committing to a real platform.

For anything else, Anthropic published a channels reference for building custom channel plugins. All channel plugins are currently Bun scripts.

The OpenClaw Elephant in the Room

The timing of this launch is not subtle. OpenClaw — the open-source autonomous AI agent created by Peter Steinberger — has been one of the most talked-about developer tools of 2026. Since going viral in late January, it has attracted enterprise adoption from companies like Tencent Cloud and Alibaba, and it essentially proved that developers want an AI agent they can message from anywhere, not just from a terminal window.

Claude Channels addresses the same core use case: persistent, multi-channel access to an AI coding agent. But the approaches differ in important ways.

Architecture. OpenClaw is a standalone agent that wraps multiple LLMs and connects to various services through plugins. Claude Channels is a feature built into Claude Code, tied to Anthropic’s infrastructure and models. OpenClaw gives you choice of model; Claude Channels locks you into Claude.

Security model. This is where Anthropic draws its sharpest contrast. OpenClaw has faced documented security issues — a Meta AI security researcher reported an OpenClaw agent “running amok” on her inbox, and the tool is known to be susceptible to prompt injection attacks. Claude Channels takes a more conservative approach: every channel plugin maintains a sender allowlist, only paired accounts can push messages, and Team/Enterprise organizations must explicitly enable channels through admin settings. Being listed in .mcp.json is not enough to push messages — a server must also be named in the --channels startup flag.

Ease of setup. OpenClaw requires self-hosting and managing your own infrastructure. Claude Channels works within the existing Claude Code environment — install a plugin, configure a token, restart with --channels, and you are live. For developers already using Claude Code, the onboarding friction is minimal.

Cost model. OpenClaw is free and open-source (you pay for LLM API calls). Claude Channels runs on your existing Claude subscription, but since it keeps a session active and Claude processes every incoming event, costs can add up. One documented case showed a developer spending $100 in a single hour when Claude was actively debugging a complex issue. Anthropic has not published specific pricing guidance for always-on channel usage.

The honest assessment: Claude Channels is not a full OpenClaw replacement. OpenClaw supports more integrations, more LLM backends, and gives users complete control. But for developers already in the Claude ecosystem who want the “message your AI from your phone” workflow without self-hosting, Claude Channels removes a lot of friction.

What the Developer Community Is Saying

The Hacker News thread surfaced several recurring themes from early adopters and skeptics.

The “why Telegram?” question. Multiple commenters were surprised Anthropic chose Telegram over Slack for its first messaging integration. The reasoning appears to be practical: Telegram has roughly 1 billion monthly active users compared to Slack’s 50 million, and its bot API is widely regarded as one of the cleanest programmatic chat integrations available. Slack support is not included in the research preview but seems like an obvious next step.

Security anxiety. Despite the allowlist model, developers expressed concern about giving an AI agent the ability to act autonomously based on external events. The permission model partially addresses this — if Claude hits a permission prompt while you are away, the session pauses until you approve locally. But for unattended use, Anthropic offers a --dangerously-skip-permissions flag (the name tells you everything about how they feel about it). Running an AI agent that can modify code based on incoming Telegram messages, with permissions bypassed, in a production environment — that makes a lot of engineers nervous.

Stability complaints. The research preview label is doing real work here. Reported issues include Remote Control websocket disconnections every 10 minutes, a GitHub connector described as “unavailable or non-functional,” and documentation gaps. This is not production-ready, and Anthropic is upfront about that — the --channels flag syntax and protocol contract may change.

The economic question. Several Hacker News commenters flagged the elephant nobody at Anthropic is talking about: if Claude Channels succeeds and developers run always-on sessions that process events around the clock, what happens to the cost model? Current Claude Pro and Max plans were designed around interactive use, not 24/7 event processing. Anthropic has not addressed how subscription tiers will adapt.

Requirements and Limitations

Before getting excited about running Claude as your always-on coding assistant, here are the constraints worth knowing:

Claude Code v2.1.80 or later is required
claude.ai login only — Console and API key authentication are not supported
Events only arrive while the session is open — there is no background daemon. You need a persistent terminal, a tmux session, or a background process to keep things running
Research preview restrictions — the --channels flag only accepts plugins from an Anthropic-maintained allowlist. Custom channels require a separate --dangerously-load-development-channels flag
Team and Enterprise organizations must have channels explicitly enabled by an admin through managed settings. It is disabled by default for these plans
No Zero Data Retention support — organizations with ZDR enabled cannot use channels

These are not minor caveats. The “session must stay open” requirement alone means Claude Channels is not truly always-on — it is “on as long as your terminal is.” That gap between marketing promise and technical reality is something Anthropic will need to close.

Where This Fits in Anthropic’s Bigger Picture

Claude Channels is not an isolated feature. It ships alongside several other recent additions to the Claude Code ecosystem: Cowork (a persistent agent thread), the Claude Agent SDK (formerly the Claude Code SDK), Remote Control (driving a local session from your phone), and Scheduled Tasks (polling on a timer). Together, these features sketch out Anthropic’s vision: Claude as a persistent, multi-modal collaborator that operates across your entire development workflow, not just inside a terminal window.

The competitive pressure is real. OpenClaw proved the market. GitHub Copilot and Cursor are pushing their own agent capabilities. Google’s Gemini CLI launched as free and open-source. Anthropic is betting that tight integration, strong security defaults, and the quality of Claude’s underlying models will win developers over — even if the ecosystem is more locked-down than the open-source alternatives.

Whether that bet pays off depends on execution. The research preview is rough around the edges, the cost model is unclear, and the always-on story has a terminal-shaped hole in it. But the core idea — an AI coding agent that listens to your world and acts on it — is clearly where the industry is heading. Anthropic wants to get there first.

FAQ

Is Claude Channels free to use?
Claude Channels is available to users on Claude Pro and Max plans through claude.ai login. There is no additional charge for the channels feature itself, but all Claude Code usage (including processing channel events) counts against your plan’s usage limits. For always-on use cases, costs can escalate quickly — monitor your usage carefully.

How does Claude Channels compare to OpenClaw?
Both let you message an AI agent from platforms like Telegram and Discord. OpenClaw is open-source, supports multiple LLM backends, and offers more integrations. Claude Channels is tightly integrated with Claude Code, offers stronger security defaults (sender allowlists, admin controls), and requires no self-hosting. OpenClaw gives more flexibility; Claude Channels gives more convenience for existing Claude users.

Can I use Claude Channels with Slack or other messaging platforms?
The research preview only supports Telegram and Discord as messaging platforms, plus a localhost demo called Fakechat. Slack is not currently supported. However, Anthropic published a channels reference for building custom channel plugins, so developers can create their own integrations for platforms not yet officially supported.

Is it safe to let Claude act on external events autonomously?
Anthropic built several safety layers: sender allowlists restrict who can push events, sessions require explicit opt-in via the --channels flag, and permission prompts pause the session when Claude needs approval for sensitive actions. For fully unattended operation, the --dangerously-skip-permissions flag exists but is not recommended for production use. The security model is conservative by design, but running any autonomous agent on a production codebase carries inherent risk.

Does Claude Channels work in the background?
Not natively. Events only arrive while a Claude Code session is open. To approximate always-on behavior, you need to run Claude Code in a persistent terminal session (such as tmux or screen) or a background process. There is no standalone daemon or server component — this is a meaningful limitation for developers expecting true 24/7 operation.

Top AI Product

Leave a comment Cancel reply