DEV Community: Ayyaz Zafar

Why Most Free Claude Code Setups Break (And the One That Doesn't)

Ayyaz Zafar — Sat, 11 Apr 2026 13:36:21 +0000

There's a specific reason why plugging a free model into Claude Code usually fails, and it's not the model quality.

The Real Requirement: Tool Calling

Claude Code's agent loop works like this:

Receive task
Call a tool (create file, run command, read output)
Observe result
Continue or adjust

Step 2 requires the model to support function calling — the ability to call structured tools, not just generate text. Most free models either don't support it at all, or implement it in a way that breaks at step 2 or 3. The agent starts working and then silently fails or returns garbage.

This is why OpenRouter free tiers, most local models, and generic API proxies don't work well with Claude Code as an agent. They work fine for chat. They fail for agentic tasks.

Gemma 4 31B: Native Function Calling

Gemma 4 31B has proper native function calling support. That's what makes it actually work as a coding agent — not just a code generator that hands text back to you.

To verify: I ran two tests.

Test 1: Build a Python terminal dashboard

Create a Python script called dashboard.py that:
- Generates sample SaaS metrics (Revenue, Users, Signups, Churn)
- Prints 4 metric cards with trend arrows
- ASCII bar chart for 6 months of revenue
- Recent transactions table
- Python standard library only, run it after creating

It created the file, ran it, verified the output. Full agent loop, zero issues.

Test 2: Find and fix 3 bugs in user_report.py

The script had:

= instead of == in a list comprehension (syntax error)
datetime.datetime.now without () (subtle runtime error)
user[email] instead of user["email"] (missing quotes)

Gemma 4 found all three, explained each one, fixed them, and ran the fixed version to confirm output.

The Setup (One Command)

Install Ollama if you don't have it:

curl -fsSL https://ollama.com/install.sh | sh

Then:

ollama launch claude --model gemma4

This connects Claude Code to Gemma 4 31B running on Ollama's cloud. The download is a small routing file — not 20GB. First run asks you to authenticate with Ollama. After that, it's instant.

Honest Limitations

Free tier: Measured in GPU time, not tokens. Resets periodically. A normal coding session fits fine. Heavy daily use will hit the limit — $20/month gives 50x more headroom.

HTML generation bug: Current version produces malformed output (doubled tags) for HTML specifically. Python, shell, JSON, config files — all fine. Don't use it for frontend templating.

When This Is Useful

If you want to test Claude Code's agentic capabilities before committing to an Anthropic plan, this is the most functional free option available right now. The tool calling works, the agent loop completes, and the model is capable enough to handle real tasks.

Full writeup: ayyaztech.com/blog/gemma4-claude-code-ollama-cloud-free

The 3 Things That Trip Up Every VPS Beginner (And How to Handle Each One)

Ayyaz Zafar — Fri, 10 Apr 2026 15:25:43 +0000

Setting up a VPS for the first time is manageable — until you hit one of three walls. Here's what they are and exactly how to get past each one.

Wall 1: UFW Firewall

Most tutorials tell you to run ufw enable and move on. The part they skip: what to allow first.

If you enable UFW before allowing SSH, you lock yourself out of your own server.

Safe sequence:

ufw allow 22      # SSH — do this FIRST
ufw allow 80      # HTTP
ufw allow 443     # HTTPS
ufw enable        # now enable it
ufw status        # verify

That's it. Everything not on this list is blocked. An open VPS gets port-scanned within hours of provisioning — don't skip this step.

Wall 2: nginx Reverse Proxy

Your Node.js app runs on port 3000. Visitors come in on port 80. nginx is the bridge between them.

The config that makes this work:

server {
    listen 80;
    server_name your-domain.com;

    location / {
        proxy_pass http://localhost:3000;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection 'upgrade';
        proxy_set_header Host $host;
        proxy_cache_bypass $http_upgrade;
    }
}

Save this to /etc/nginx/sites-available/myapp, symlink it to sites-enabled, test with nginx -t, reload with systemctl reload nginx. That's the full flow.

Wall 3: PM2 (The One Everyone Forgets)

If you start your app with node app.js and close the terminal — app is gone. If the server reboots — app is gone.

PM2 fixes this:

npm install -g pm2
pm2 start app.js --name myapp
pm2 startup        # generates a command — run it
pm2 save           # saves the process list for reboot

Two commands most people miss: pm2 startup and pm2 save. Without both, your app won't survive a reboot.

Putting It All Together with Claude Code

If you want to see these three things handled automatically on a fresh VPS, I put together a full walkthrough where Claude Code does the entire setup from one SSH prompt — including the firewall, reverse proxy, and PM2 config.

Full guide: ayyaztech.com/blog/hostinger-vps-setup-claude-code

Video walkthrough: youtu.be/AX9Wp_YxTpU

Claude AI in Slack: Summarize Threads, Review Code, and Fix Bugs with a PR

Ayyaz Zafar — Thu, 09 Apr 2026 19:28:40 +0000

What if you could ask Claude to fix all the bugs in your GitHub repo — without leaving Slack?

One message. It clones the repo, finds the bugs, fixes them, and opens a pull request. Here's the full setup and three real demos.

What You'll Be Able to Do

✅ Summarize threads and extract action items instantly
✅ Review code snippets pasted directly in a channel
✅ Analyze your entire GitHub repo, fix bugs, and create a PR — all from one message

Prerequisites

A Slack workspace where you have permission to install apps
A Claude Pro or Max plan (Claude Code routing requires this)
At least one GitHub repo connected on claude.ai/code

Step 1: Install Claude in Your Slack Workspace

Open Slack → More → Tools → Apps → Search "Claude" → Add to Slack → Authorize.

Once installed, the Claude app appears in your Slack sidebar.

Step 2: Connect Your Claude Account

Click the Claude app → connect your account via the App Home tab.

Step 3: Configure Routing Mode (Critical)

Go to the Home tab of the Claude app. Set Routing Mode to Code + Chat.

Without this, Claude won't route to Claude Code for repository work.

Step 4: Connect Your GitHub Repo

Go to claude.ai/code, click Select Repository, and connect the repo you want Claude to work with.

Demo 1: Summarize a Thread in 5 Seconds

In any Slack channel (invite Claude first with /invite @Claude):

@Claude summarize this thread and list any action items

5 seconds. Full summary with action items extracted. Ideal for catching up after being away.

Demo 2: Code Review Without Leaving Slack

Paste code directly in Slack:

router.get('/:id', (req, res) => {
  const user = users.find(u => u.id === req.params.id);
  res.json(user);
});

Then ask:

@Claude review this code — why might it return undefined for a valid user ID?

Claude catches it immediately: req.params.id is a string, but user IDs are numbers. Strict === comparison always fails. No tab switching needed.

Demo 3: Fix Your Entire Repo from Slack

@Claude look at this repo: https://github.com/your-username/your-repo
Find all the bugs and fix them.

Claude:

Routes to Claude Code automatically
Clones and reads all files
Lists every bug (location + explanation)
Fixes them
Creates a branch and opens a PR

You get two buttons: View session (full Claude Code web view) and Create PR (one click, PR is live).

In my demo it found 4 bugs: string/number comparison in GET and DELETE, missing validation on POST, and broken ID generation.

Important Limitations

No session memory across messages — each new @claude mention = fresh session. For follow-up work, use the Claude Code web session it opens.
Channels only — doesn't work in DMs.
Requires Pro or Max plan + GitHub connected.

Quick Setup Checklist

[ ] Claude installed in Slack workspace
[ ] Account connected (App Home → Connect)
[ ] Routing Mode → Code + Chat
[ ] GitHub repo connected on claude.ai/code
[ ] Claude invited to your channel

The Slack → Claude Code → GitHub PR flow is what makes this worth setting up. One prompt, full diff, ready to review.

STOP — Claim Your Free Claude Credits Before April 17 (New Usage Bundles)

Ayyaz Zafar — Mon, 06 Apr 2026 16:36:26 +0000

Anthropic just changed how third-party tools work with your Claude subscription. Tools like OpenClaw now use "extra usage" instead of your plan.

To make up for it, they're giving every subscriber a one-time FREE credit:
💰 Pro plan → $20 free
💰 Max 5x plan → $100 free
💰 Max 20x plan → $200 free

I claimed my $100 live in this video. Took 5 seconds.

⚠️ You MUST claim before April 17, 2026 or you lose it.

They also launched usage bundles:
• $50 bundle → pay $45 (10% off)
• $250 bundle → pay $200 (20% off)
• $1000 bundle → pay $700 (30% off)

Claim here: https://claude.ai/settings/usage
Full walkthrough: https://youtu.be/rfTmTpAcPvs

Deploy Next.js to Hostinger with One AI Prompt (Claude Code Tutorial)

Ayyaz Zafar — Sun, 05 Apr 2026 17:20:13 +0000

What if deploying a Next.js app to a VPS took just one sentence?

You just bought a fresh Hostinger VPS. It's a blank Ubuntu server — nothing installed. No Node.js, no Nginx, no PM2, nothing.

Normally, you'd spend 30+ minutes SSHing in, installing packages, creating config files, setting up reverse proxies, and troubleshooting why port 80 isn't working.

What if you could skip all of that and just tell an AI to do it?

That's exactly what this tutorial shows. One prompt to Claude Code, and your Next.js app is live on your VPS.

Watch the Full Tutorial

What You Need

A Hostinger VPS running Ubuntu (fresh, nothing installed)
Claude Code installed on your Mac or Linux machine
Your VPS IP address from the Hostinger dashboard

Step 1: Set Up SSH Key Access

Before Claude Code can connect to your server, you need SSH key authentication. This is a one-time step.

Generate an SSH Key (if you don't have one):

ssh-keygen

Press Enter through the prompts to accept defaults. This creates a key pair in ~/.ssh/.

Copy your public key:

cat ~/.ssh/id_rsa.pub

Copy the entire output and paste it into your Hostinger VPS dashboard under SSH Keys.

Step 2: The One Prompt That Does Everything

Open Claude Code in your terminal and give it this single prompt:

SSH into my server at root@YOUR_SERVER_IP, install Node.js, 
create a new Next.js app at /var/www/mysite, configure nginx 
as a reverse proxy to port 3000, set up a systemd service to 
keep the app running, and make it live on port 80.

Replace YOUR_SERVER_IP with your actual Hostinger VPS IP address.

That's it. One prompt.

What Claude Code Does Automatically

After you hit Enter, Claude Code handles everything — without you typing a single SSH command:

SSHs into your server using your key
Updates system packages
Installs Node.js (via NodeSource)
Creates a Next.js app at /var/www/mysite
Builds the app for production
Installs and configures Nginx as a reverse proxy to port 3000
Creates a systemd service so the app auto-starts on reboot
Starts the app and makes it live on port 80

The whole process takes a few minutes — compared to 30+ minutes of manual configuration.

See It Live

Once Claude Code finishes, open your browser and navigate to your VPS IP address. You'll see your Next.js app running live — the default welcome page, served via Nginx on port 80.

No Docker. No CI/CD pipeline. No YAML files. Just one prompt.

Bonus: Update Your App with a Follow-Up Prompt

Want to make changes? Just give Claude Code another prompt:

SSH into my server at root@YOUR_SERVER_IP and change the heading 
of the Next.js app at /var/www/mysite to "Deployed by AI", 
then rebuild and restart the app.

Claude Code will SSH in, find the right file, update the heading, rebuild the app, restart the systemd service, and your site is updated.

This Works for More Than Next.js

The same pattern works for deploying:

React apps (with serve or Nginx for static files)
Angular apps
Python/Django/Flask apps
Express.js APIs
Any web application that runs on a Linux server

Just describe what you want in plain English, and Claude Code figures out the commands.

Why Hostinger VPS?

Hostinger VPS is an excellent choice for this workflow:

Affordable pricing — plans start very low for a full Ubuntu VPS
Clean Ubuntu installs — fresh slate, no bloatware
SSH key support built into the dashboard
Fast provisioning — server ready in minutes
Full root access — install whatever you need

Conclusion

Server deployment doesn't have to be tedious and error-prone. With Claude Code and a Hostinger VPS, you can go from a blank server to a live Next.js application in one prompt.

Try it yourself: grab a Hostinger VPS, install Claude Code, and deploy your next project with a single sentence.

Open Code + MiniMax M2.7: Get 95% Cheaper AI Coding (Claude Code Alternative)

Ayyaz Zafar — Sat, 04 Apr 2026 13:38:22 +0000

How I replaced my $200/month Claude Code subscription with a setup that costs $5-15/month — and barely noticed the difference.

If you've been using Claude Code (or want to), you know the pain: it's incredible, but it's expensive. Claude Code Max runs $100-200/month, and even API usage with Claude Opus 4.6 at $5/M input and $25/M output adds up fast.

What if I told you there's an open-source alternative with 135K+ GitHub stars that gives you 90% of Opus quality at roughly 95% less cost?

Meet Open Code + MiniMax M2.7 — the combo that changed my workflow completely.

Full blog post: Read on AyyazTech

Why Open Code?

Open Code is an open-source AI coding assistant that runs in your terminal — just like Claude Code. But here's the key difference: you can use any AI model with it, not just Claude. That means you can pick cheaper models that still deliver great results.

With 135K+ GitHub stars, it's one of the most popular AI coding tools out there.

Why MiniMax M2.7?

MiniMax M2.7 is where the magic happens. According to Kilo Code benchmarks, it delivers approximately 90% of Claude Opus quality. But look at the pricing difference:

Cost Comparison

	MiniMax M2.7	Claude Opus 4.6
Input	$0.30/M tokens	$5.00/M tokens
Output	$1.20/M tokens	$25.00/M tokens
Monthly (typical)	$5-15	$100-200

That's roughly a 95% cost reduction for 90% of the quality. For most day-to-day coding tasks, you won't notice the difference.

Setup Guide (5 Minutes)

Step 1: Install Open Code

curl -fsSL https://opencode.ai/install | bash

Step 2: Authenticate via OpenRouter

We use OpenRouter as the provider because MiniMax direct requires a $25 minimum topup. OpenRouter has no minimum — even $1 works.

opencode auth login

This opens an interactive CLI. Select OpenRouter from the provider list, then paste your OpenRouter API key when prompted.

Step 3: Select MiniMax M2.7

Launch Open Code:

opencode

Once inside, type:

/models

Search for MiniMax M2.7 and select it. That's it — you're ready to code.

What Can It Do? (Three Live Demos)

In the video, I put this setup through three real-world tests:

Demo 1: Node.js Express API with JWT Auth

I asked it to build a complete REST API with signup, login, and a protected route using JWT authentication. It generated clean, working code on the first try — proper middleware, token handling, everything.

Demo 2: Python Bug Fix

I gave it a buggy Python function for merging sorted arrays that was returning incomplete results. It identified the issue, explained what was wrong, and fixed it correctly.

Demo 3: Responsive Landing Page with Tailwind CSS

I asked for a full landing page with a dark theme, hero section, and feature cards using Tailwind CSS. It produced a polished, responsive design that looked production-ready.

When to Use This vs Claude Code

Use Open Code + MiniMax M2.7 for:

Day-to-day coding tasks
Building features and prototypes
Bug fixes and debugging
Learning and experimentation
When budget matters

Keep Claude Code for:

Complex architectural decisions
Very large codebase refactoring
When you need absolute top-tier reasoning

The Bottom Line

You don't need to spend $200/month to get high-quality AI coding assistance. Open Code + MiniMax M2.7 through OpenRouter gives you an incredible developer experience at a fraction of the cost.

Try it today — the setup takes 5 minutes, and your wallet will thank you.

Watch the full tutorial with live demos: YouTube Video

Read the detailed blog post: AyyazTech Blog

Run Claude Code with a Free Local Model — Qwen 3.5 + Ollama Setup

Ayyaz Zafar — Thu, 02 Apr 2026 08:52:22 +0000

Claude Code is powerful but costs money. Every prompt burns API tokens and your code is sent to external servers. What if you could run the same workflow with a free local model?

Meet Qwen 3.5

A 27B parameter model distilled from Claude 4.6 Opus reasoning traces:

Beats Claude Sonnet 4.5 on SWE-bench
96.91% HumanEval accuracy
24% less chain-of-thought bloat (faster responses)
Fits on a single GPU with 16GB VRAM
300K+ downloads on Hugging Face

Setup

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull the model
ollama pull qwen3.5

# Launch Claude Code with local model
claude --model ollama:qwen3.5

One command and you're coding with local AI. Same workflow as Claude, running on your GPU.

What It Can Do

Write code: Clean async Python with error handling and type hints
Fix bugs: Reads file, finds bug, explains why, fixes in place
Build UIs: Full dark-theme landing page with Tailwind CSS from one prompt

All locally. No API call. No internet required.

Why This Matters

Free forever — no API costs, no rate limits
Private — your code never leaves your machine
Same workflow — identical Claude Code experience
Opus-level reasoning — distilled from Claude 4.6 Opus

Qwen 3.5 + Ollama + Claude Code = full agentic AI coding, running locally, free forever.

Originally published at ayyaztech.com

Claude Code Built My Portfolio in 5 Minutes — Next.js + Tailwind CSS from One Prompt

Ayyaz Zafar — Wed, 01 Apr 2026 12:45:14 +0000

What if you could build an entire developer portfolio — Next.js, Tailwind CSS, fully responsive, dark mode, animations — from a single prompt? In under 5 minutes?

I started with an empty folder. No boilerplate, no templates. Just Claude Code and one detailed prompt.

What Claude Code Built

One prompt created:

Navbar — responsive with animated mobile menu
Hero — clean, bold landing section
About — hover effects, professional styling
Projects — card layout with hover animations
Skills — each skill changes color with animation
Contact — form with clean layout, fully responsive

All with TypeScript, Tailwind CSS, and the Next.js App Router.

The Real Power: Iterating

Building it is impressive. But changing it is where AI shines.

I asked: "Add a floating navigation bar that changes background on scroll. Also add a dark/light mode toggle button."

One sentence. Claude Code reviewed existing files, edited components, and delivered:

Floating navbar — semi-transparent, changes color on scroll
Dark/light mode toggle — all colors update properly in both modes

The Numbers

Manual development: 2-4 hours
Claude Code: Under 5 minutes

No templates. No boilerplate. Just one prompt and iterative refinement.

The key is prompt quality — a detailed prompt describing sections, features, and design expectations gives Claude Code enough context to make good architectural decisions.

Originally published at ayyaztech.com

I Control My Server from Telegram with AI — OpenClaw Setup Guide

Ayyaz Zafar — Tue, 31 Mar 2026 19:42:19 +0000

If you run a VPS, you know the drill. Something feels off, you open a terminal, SSH in, try to remember the right command, check the logs, restart a service. And if you're on your phone? Forget it.

What if you could just text your server and ask it what's going on — in plain English? That's exactly what we're building today with OpenClaw, a free, open-source AI agent that lives on your server and talks to you through Telegram.

What You'll Build

By the end of this guide, you'll have an AI agent on your server that you can message from Telegram to:

Check if your website is up and healthy
Monitor disk space, CPU, and memory usage
Read and summarize nginx access logs
Restart services like nginx or your Node.js app
Edit source code and deploy changes — all from one message
Diagnose performance issues with AI-powered analysis

The Architecture

[VPS 1: Website]  <--- SSH --->  [VPS 2: OpenClaw + Telegram]
   nginx                              AI Agent
   Next.js app                        Controls VPS 1

VPS 1 runs your website (a Next.js app behind nginx). VPS 2 runs OpenClaw, which connects to Telegram and has SSH access to VPS 1.

Step 1: Create a Dedicated User

SSH into your second VPS. Create a dedicated user — don't run this as root:

adduser openclaw
usermod -aG sudo openclaw
loginctl enable-linger openclaw
su - openclaw

Step 2: Install OpenClaw

curl -fsSL https://openclaw.ai/install.sh | bash

The installer starts onboarding automatically. Connect your LLM (OpenRouter + DeepSeek V3.2), then connect Telegram via BotFather bot token.

Step 3: Set Up the Background Service

source ~/.bashrc
mkdir -p /run/user/$(id -u)
echo 'export XDG_RUNTIME_DIR=/run/user/$(id -u)' >> ~/.bashrc
echo 'export DBUS_SESSION_BUS_ADDRESS="unix:path=${XDG_RUNTIME_DIR}/bus"' >> ~/.bashrc
source ~/.bashrc
openclaw gateway install --force
systemctl --user daemon-reload
systemctl --user start openclaw-gateway.service
systemctl --user enable openclaw-gateway.service

Verify: openclaw gateway status — should show "Runtime: running".

Step 4: SSH Access to Website Server

ssh-keygen -t ed25519
ssh-copy-id root@YOUR_WEBSITE_SERVER_IP

Step 5: Pair with Telegram

Send your bot a message, then approve the pairing code:

openclaw pairing approve telegram YOUR_CODE

Step 6: Give OpenClaw Context

Send one detailed message in Telegram with your server's IP, SSH user, file paths, service names, and deploy steps. OpenClaw has persistent memory — it remembers everything.

What It Can Do

Health Check: "Is my site up?" → SSHes in, curls the site, reports status
Disk Space: "How much disk space is left?" → Human-readable summary
Log Analysis: "Show nginx access logs from last hour" → AI-summarized traffic patterns
Service Restart: "Restart nginx" → Done. Confirmed.
Code + Deploy: "Change the heading to 'Powered by AI'" → Edits code, rebuilds, restarts service
AI Diagnosis: "My site feels slow" → Checks CPU, memory, disk, nginx, response time — full diagnosis

Why Not Just a Bash Script?

A script runs one command. OpenClaw thinks. It checks multiple metrics, synthesizes a diagnosis, remembers your server config, and gives human-readable answers instead of raw terminal output.

OpenClaw is free, open-source, and runs on any server. Connect it to Telegram, Discord, WhatsApp, or whatever you use.

Originally published at ayyaztech.com

Claude Code Channels: Text Your AI from iMessage — Complete Setup Guide

Ayyaz Zafar — Mon, 30 Mar 2026 16:12:22 +0000

Claude Code Channels now supports iMessage natively on Mac. You can text your AI assistant from your iPhone and it texts you back — no Telegram bot, no Discord server, no external service. Just iMessage.

If you've seen my Discord and Telegram channel guides, this is the simplest one yet.

How It Works

Your Mac stores all iMessage data in a SQLite database called chat.db. The iMessage plugin polls this database every second for new messages. When it finds one, it pushes the message to Claude Code. For replies, Claude tells Messages.app to send a text through AppleScript.

Everything stays local on your Mac. No cloud services, no API tokens, no hosted bots.

Prerequisites

macOS (iMessage is Mac-only)
Claude Code installed and up to date
Full Disk Access permission for your terminal app

Step 1: Grant Full Disk Access

macOS protects the chat.db database. Your terminal needs Full Disk Access to read it.

Open System Settings
Go to Privacy & Security → Full Disk Access
Add your terminal app (Terminal, iTerm2, Warp, or VS Code)
Restart your terminal if prompted

The key rule: whatever app launches bun (which runs Claude Code) needs this permission. Without it, the plugin exits immediately.

Step 2: Install the iMessage Plugin

First, make sure Claude Code is up to date:

claude update

Install the iMessage plugin:

/plugin install @anthropic/claude-plugin-imessage

If you get a "plugin not found" error, install or update the official marketplace first:

/plugin marketplace add anthropic/claude-plugin-official

Choose user scope when prompted so the plugin works across all your projects.

Step 3: Launch Claude Code with Channels

Exit Claude Code and restart with the channels flag:

claude --channels plugin:imessage:imessage

You should see: "Listening for channel messages". That's it — the plugin is now polling your messages database.

Step 4: Test with Self-Chat

Open Messages on your iPhone and text yourself (your iCloud phone number or email). Self-chat bypasses all access control — it works immediately with no setup.

Try sending: "Hey Claude, what's in my working directory?"

Note: In self-chat, every message appears twice (as sender and receiver). That's normal iMessage behavior, not a bug. To avoid this, allow-list a second phone number and text from that device instead.

Step 5: Access Control

By default, only your self-chat works. Everyone else is silently dropped. To let someone else reach your Claude Code session:

/imessage:access allow +1234567890
/imessage:access allow friend@icloud.com

Each time you add someone, Claude Code asks for confirmation — you can approve from your terminal OR directly from your iPhone by typing the secret code.

To remove access:

/imessage:access remove +1234567890

Real Use Cases

Remote file lookup: Text Claude to check config values while away from desk
Build things remotely: Text "Create a landing page" — file is ready when you return
Check build status: Text "Did the tests pass?" — get results on your phone
Standup prep: Text "What did I commit yesterday?" — formatted summary on iPhone

Additional Configuration

Disable the Signature

IMESSAGE_APPEND_SIGNATURE=false claude --channels plugin:imessage:imessage

Group Chat Support

/imessage:access set mention-patterns ["@claude", "hey claude"]

Bypass Permissions (Use Carefully)

claude --channels plugin:imessage:imessage --dangerously-skip-permissions

Things to Know

Claude Code must be running — close the terminal, messages stop
macOS only — chat.db only exists on Mac
First reply triggers an automation prompt — click OK (one-time)
Security matters — this reads your personal messages database

Related Guides

All three channel options work. Pick whichever platform you use the most.

Claude Code Channels: Control Your AI Agent from Telegram (Complete Setup Guide)

Ayyaz Zafar — Mon, 23 Mar 2026 16:08:22 +0000

Control Claude Code from Your Phone via Telegram

What if you could message your AI coding agent from your phone and have it build things on your computer — even when you're not at your desk? That's exactly what Claude Code Channels does with Telegram.

In this guide, I walk you through setting up a Telegram bot that connects directly to Claude Code. You send a message, Claude does the work, and sends the result right back to your Telegram — including screenshots of what it built.

If you saw my Discord version of this setup, this is the Telegram companion.

What You Need

Claude Code installed and working
A Telegram account
Terminal access

Step 1: Create a Telegram Bot with BotFather

Open Telegram and search for @botfather:

Click Start
Type /newbot
Enter a display name and a username ending in "bot"
Copy the bot token BotFather gives you

Step 2: Install the Telegram Plugin

claude update
claude plugin install @anthropic/channel-telegram

Choose global scope. If you get "plugin not found":

claude marketplace refresh
# or first-time setup:
claude marketplace install @anthropic/official

Then reload and restart your Claude Code session (important!):

claude plugin reload

Step 3: Configure and Pair

/telegram:configure   # paste your bot token
claude --channel telegram   # start listening

In Telegram: message your bot → get a 6-character pairing code → enter it in the terminal → approve.

Step 4: Lock Down Access

/telegram:access

Change DM policy from pairing to allow-list. Only approved senders can now interact with your bot.

Live Demo

I asked Claude Code from Telegram to build a landing page. It created the HTML, opened it in my browser, took a screenshot, and sent it back to Telegram. All remotely.

Unattended Mode

For fully remote control without permission prompts:

claude --channel telegram --dangerously-skip-permissions

Only use in trusted environments. This bypasses all safety prompts.

Use Cases

Build web pages and get screenshots on your phone
Organize files and folders remotely
Get YouTube transcript summaries on the go
Run development tasks from anywhere

Claude Code Channels: Control Your AI Agent from Discord (Complete Setup Guide)

Ayyaz Zafar — Sat, 21 Mar 2026 14:00:06 +0000

Developers were paying for Mac Minis just to run OpenClaw so they could message their AI agent from their phone. Anthropic just made that completely unnecessary. Claude Code Channels is the native, secure version of what OpenClaw was trying to do — and it supports Discord and Telegram out of the box.

In this guide, I'll walk you through the complete setup process: from testing with a local Fake Chat demo, to creating a Discord bot, pairing your account, and locking down access for security.

What Are Claude Code Channels?

A channel is an MCP server that pushes events into your running Claude Code session. This means Claude can react to things that happen while you're not at the terminal. Channels are two-way — Claude reads the event and replies back through the same channel, like a chat bridge.

Key points:

Events only arrive while the session is open
For an always-on setup, run Claude in a background process or persistent terminal
Discord and Telegram are included in the research preview
You can build your own custom channels for Slack, WhatsApp, or anything else

Quick Start: Testing with Fake Chat

Before connecting Discord, let's verify everything works with the built-in Fake Chat demo — a local chat UI that requires no authentication or external services.

# Make sure Claude Code is up to date
claude update

# Add the official plugins marketplace (if not already added)
claude plugin add-marketplace claude-plugins-official

# Install the Fake Chat plugin
claude plugin install @anthropic-ai/claude-code-fakechat

# Start Claude Code with the Fake Chat channel
claude --channels fakechat

Once running, open http://localhost:8787 in your browser. You'll see a simple chat interface where you can type messages and even attach files.

Setting Up Discord

Step 1: Create a Discord Bot

Go to the Discord Developer Portal
Click New Application and name it (e.g., "Claude Code")
Go to the Bot section in the left sidebar
Click Reset Token and copy the token — save it somewhere safe
Enable Message Content Intent under Privileged Gateway Intents
Click Save Changes

Step 2: Invite the Bot to Your Server

Go to OAuth2 → URL Generator
Select the bot scope
Under Bot Permissions, enable: View Channels, Send Messages, Read Message History, and Attach Files
Copy the generated URL and open it in a new tab
Select your server and click Authorize

Step 3: Install and Configure the Discord Plugin

# Install the Discord channel plugin
claude plugin install @claude-plugins-official/discord

# Configure with your bot token
/discord:configure YOUR_BOT_TOKEN_HERE

# Restart Claude Code with the Discord channel enabled
claude --channels discord

Step 4: Pair Your Discord Account

DM your bot on Discord — it will reply with a pairing code
Run the pairing command shown in the bot's reply in your Claude Code terminal
Approve the pairing when prompted
You're connected! Send messages from Discord and Claude responds

Security: Locking Down Access

This step is critical. Without locking down access, anyone who can message your bot could potentially control your Claude Code session — and by extension, your entire computer.

# Lock down to allow-list mode
/discord:access policy allowlist

This changes the DM policy from pairing (anyone can request access) to allowlist (only pre-approved users can interact).

What Can You Do from Discord?

Once connected, you can do everything from Discord that you'd normally do in the terminal:

Send text messages — ask Claude to read files, explain code, or answer questions
Upload images — Claude can analyze screenshots, diagrams, or any visual content
Create files — ask Claude to generate code, HTML pages, configs, etc.
Run commands — anything Claude Code can do locally, you can trigger remotely

Bypass Permissions (Use with Caution)

# WARNING: Only use in sandboxed/trusted environments
claude --channels discord --dangerously-skip-permissions

⚠️ Warning: This flag bypasses ALL permission prompts. Only use this in sandboxed environments you can easily restore.