← All articles

Coding

Claude Code 2026 Updates: Channels, Remote Control, and /loop Scheduling

7 min read

Updated

Explore the March 2026 updates to Claude Code, featuring native Slack/Telegram Channels, Remote Control, Dispatch, and automated /loop scheduling.

March 2026 brought updates that significantly expand the capabilities of Claude Code.

Previously a powerful terminal-based agentic coding tool, the platform has introduced capabilities that allow you to control sessions from your phone, schedule autonomous runs, and message agents directly through platforms like Telegram and Discord.

These features—including Channels, Remote Control, Dispatch, and /loop scheduling—provide native alternatives to third-party tools like OpenClaw, allowing developers to manage workflows directly within Claude Code without additional subscription costs.

Below is a breakdown of what was released, how it works, and how to apply it to your development workflows.

What Shipped in March 2026

Claude Code releases updates on a fast, weekly schedule. The key updates from March 2026, based on official logs, include:

DateVersionFeatureImpact
Mar 5v2.1.76Enhanced MCP server supportBetter plugin ecosystem for custom integrations
Mar 10v2.1.78Remote Control (Pro rollout)Control terminal sessions from phone/web
Mar 14v2.1.80Auto Mode safety layersSafer autonomous execution with guardrails
Mar 19v2.1.82Channels (Telegram + Discord)Message running sessions via chat apps
Mar 22v2.1.83/loop schedulingCron-style recurring autonomous agents
Mar 24v2.1.84Dispatch integrationSend tasks to agents from mobile/web
Mar 26v2.1.85Windows PowerShell supportFull Windows native experience
Mar 27v2.1.86Opus 4.6 output doublingLonger, more complete responses

The three updates that impact day-to-day workflow the most are Channels, Remote Control & Dispatch, and /loop scheduling.


Deep Dive: Channels (Telegram and Discord)

What It Is

Channels connect running Claude Code sessions directly to Telegram or Discord. By initializing a session in your terminal and linking a channel, you can send instructions and receive updates from your agent remotely.

For example, sending a message like "Run the test suite and fix any failures" via Telegram will prompt your local Claude Code instance to perform the action and report the results back in the chat.

How It Works

The architecture utilizes an MCP-based relay that bridges your messaging platform of choice with your local session:

  1. Start a Claude Code session with channels enabled.
  2. Link your Telegram bot token or Discord webhook.
  3. Send text commands from your mobile device or messaging client.
  4. Claude Code processes the instructions locally and sends status updates back to the chat.

Your source code remains entirely on your local machine; only high-level instructions and status updates travel through the messaging channel.

What You Can Do Through Channels

Channels support rich interaction, allowing you to:

  • Approve or reject changes: Respond with a simple "yes" or "no" when prompted for permissions (e.g., deleting a file).
  • Redirect tasks: Pivot the agent mid-task (e.g., "Pause that, priority bug in checkout flow").
  • Codebase queries: Ask questions like "What is the current test coverage?" to get summaries based on your project's current state.
  • Review diffs: View summaries of code changes before approving deployments or merges.

Security Considerations

Because Channels use an MCP relay architecture, your actual source files and environment variables are not sent to Telegram or Discord servers. However, because anyone with access to your bot token can command the agent, you must secure your channel credentials like private SSH keys.

Setup Example

# Start Claude Code with a Telegram channel
claude --channel telegram

# Or connect with Discord
claude --channel discord

# Configure via MCP settings
claude mcp add telegram-channel

Deep Dive: Remote Control and Dispatch

Remote Control

Remote Control lets you drive your terminal-based Claude Code session directly from a web browser or mobile phone interface. While Channels are optimized for conversational text messages, Remote Control gives you a full, responsive terminal-like interface on your phone.

# Enable remote control for the current session
claude remote-control

This generates a secure URL that you can open on any device to monitor and interact with your terminal session in real-time.

Dispatch

Dispatch is an asynchronous tool. Instead of steering an active, open session, Dispatch allows you to queue tasks from your phone or web app. Your local Claude Code agent pulls these queued tasks and processes them sequentially in the background.

Comparison: Remote Control vs. Dispatch

  • Remote Control: Best when you need to actively guide, debug, or steer a running terminal session away from your desk.
  • Dispatch: Best for queuing up isolated, one-off tasks (e.g., "Generate unit tests for auth.py") to run in the background.

Deep Dive: /loop Scheduling

The /loop command enables cron-style automation for your development environment. You specify a task and an interval, and the agent executes the instruction autonomously:

# Run tests every 30 minutes, commit fixes on pass
/loop 30m "Run the test suite, fix any failures, commit fixes"

# Run a daily check for package updates
/loop 24h "Check for outdated npm packages, update non-breaking ones, run tests"

# Check log outputs hourly
/loop 1h "Check application logs for errors, summarize any new issues"

Unlike basic cron scripts, /loop tasks run with full codebase context, allowing the agent to read files, run tests, write code, and commit changes intelligently.

Best Practices for Scheduling Loops

  • Rate Limits: Each loop iteration counts as a new prompt context. Running loops at very high frequencies (e.g., every 5 minutes) can deplete your API or subscription limits quickly. Hourly or daily intervals are recommended.
  • System Sleep: The agent runs on your local machine. If your computer goes to sleep or loses internet, active /loop tasks will pause. Consider using a dedicated server or a keep-alive script for persistent execution.
  • Human-in-the-Loop: Combine /loop with Channels. For example, configure a task to monitor logs hourly, and set it to ping your Telegram channel if it encounters an error it cannot safely resolve on its own.

Auto Mode Safety Layers

To mitigate risks associated with autonomous loops and remote access, updates include dedicated safety controls. Auto Mode requires explicit confirmation before executing destructive commands, such as:

  • File deletions outside target directories.
  • Git force pushes (git push --force).
  • Deployments to production environments.

You can customize these confirmation thresholds in your local configuration files to match your risk tolerance.


Feature Comparison Matrix

FeatureChannelsRemote ControlDispatchOpenClaw
Primary UseConversation via Telegram/DiscordRemote terminal interfaceAsync task queuingThird-party agent management
Execution LocationLocalLocalLocalVaries
Subscription CostIncluded in Pro+Included in Pro+Included in Pro+Separate service
Setup OverheadLow (Command-line config)Low (Single command)Low (Built-in)Medium-High
Context RetentionFull session contextFull session contextNew context per taskConfigurable

Claude Code vs. OpenClaw

For developers evaluating their workflow tooling, here is a direct comparison between native Claude Code features and OpenClaw:

Where Claude Code Wins

  • No Extra Fees: Included directly in standard developer plans.
  • Native Context Integration: Sessions maintain history, terminal state, and local project indexes natively.
  • Simplified Security: Code remains locally on your machine at all times.

Where OpenClaw Retains an Edge

  • Multi-Model Workflows: Supports orchestrating tasks across different provider models (e.g., Gemini, GPT, Claude) simultaneously.
  • Complex Multi-Project Views: Better suited for managing multiple unrelated codebases at the same time from a centralized server dashboard.

Pricing and Plans

All native automation features (Channels, Remote Control, /loop) are included in the standard Claude pricing tiers at no additional charge.

PlanPrice (Monthly)Price (Annual)Key Limits
Pro$20/mo$17/mo (billed annually)Standard usage quota
Max$30+/moVariesHigher rate limits, priority access
Team$150/user/moCustomAdministrative controls, SSO

Getting Started

To access these features, ensure your local CLI installation is updated:

# Update the CLI tool
npm install -g @anthropic-ai/claude-code

# Launch a session
claude

# Expose remote control
claude remote-control

# Start with channel integration
claude --channel telegram