OpenCode Server Launcher

Simple commands and scripts to launch and manage OpenCode servers, from single instances to multi-server swarms.

UNIVERSAL CAPABILITY: This skill launches servers for ANY agent type, not just the examples shown. The agent folder names (code-analyzer, documentation-writer, etc.) used in examples are for illustration ONLY. Real swarms will have completely different agent types and purposes. This skill works universally with any folder structure and any agent specialization.

Quick Start

IMPORTANT: Create custom agents in .opencode/agent/ BEFORE launching the server. Agents are only loaded when the server starts.

🚫 CRITICAL MISTAKES TO AVOID

❌ NEVER Use Timeouts on Servers or API Calls

WRONG: timeout 300s opencode serve --port 3001 & WRONG: timeout 300s curl http://localhost:3000/session/... RIGHT: opencode serve --port 3001 & RIGHT: curl http://localhost:3000/session/... (let complete naturally)

• Servers should run indefinitely, never timeout server processes
• API calls should complete naturally, don't timeout orchestrator/agent communication
• Timeouts should only be used for YOUR personal waiting periods, not for system processes

❌ NEVER Use Logging/Redirection

WRONG: opencode serve --port 3001 > /tmp/server.log 2>&1 & RIGHT: opencode serve --port 3001 &

• No logging, no redirection, no output capture
• Keep server launch commands simple and clean
• Servers run in background without complex I/O handling

❌ NEVER Use Complex Bash Commands

WRONG: cd folder && opencode serve --port 3001 > log.txt 2>&1 & PID=$!; echo "$PID" >> file RIGHT: cd folder && opencode serve --port 3001 &

• Simple background launches only
• No chaining multiple commands with semicolons
• No variable assignments in server launch commands

Basic Server Launch

# Start server on random available port
opencode serve

# Start server on specific port
opencode serve --port 3000

# Start server on all interfaces
opencode serve --port 3000 --hostname 0.0.0.0

Verify Server is Running

# Check server status
curl http://localhost:3000/config

# View available agents
curl http://localhost:3000/agent | jq .

# Check API documentation
curl http://localhost:3000/doc

Multi-Server Setup

Launch Multiple Servers

# Server 1 - General purpose
opencode serve --port 3001 &
SERVER1_PID=$!

# Server 2 - Build focused
opencode serve --port 3002 &
SERVER2_PID=$!

# Server 3 - Documentation focused
opencode serve --port 3003 &
SERVER3_PID=$!

echo "Started servers: $SERVER1_PID, $SERVER2_PID, $SERVER3_PID"

Folder-Specific Server Launch (Swarm-Ready)

For swarm deployment, launch servers in specific folders:

⚠️ EXAMPLE PATTERN ONLY - The folder names below are EXAMPLES. Use ANY folder names for your actual swarm:

# EXAMPLE: Launch server in [ANY_AGENT_FOLDER] folder
cd [agent_folder_name]
opencode serve --port 3001 &
SERVER1_PID=$!
cd ..

# EXAMPLE: Launch server in [DIFFERENT_AGENT_FOLDER] folder
cd [another_agent_folder]
opencode serve --port 3002 &
SERVER2_PID=$!
cd ..

# EXAMPLE: Launch server in [THIRD_AGENT_FOLDER] folder
cd [third_agent_folder]
opencode serve --port 3003 &
SERVER3_PID=$!
cd ..

UNIVERSAL TRUTH: The pattern works for ANY folder names:

• marketing-specialist/ → cd marketing-specialist
• legal-advisor/ → cd legal-advisor
• game-developer/ → cd game-developer
• research-scientist/ → cd research-scientist
• music-composer/ → cd music-composer
• personal-trainer/ → cd personal-trainer

REAL SWARMS WILL HAVE VASTLY DIFFERENT FOLDER NAMES - These examples only show the launching pattern.

Process Management:

• Store PIDs for later management: echo "3001:[folder_name]:$SERVER1_PID" >> .opencode/server-pids.txt
• Stop servers: kill $SERVER1_PID
• Check if running: kill -0 $SERVER1_PID

NOTE: Replace [folder_name] with your actual folder names. The pattern port:folder:pid works for ANY folder names.

Server Configuration

Environment Variables

# Set log level
export OPENCODE_LOG_LEVEL=debug

# Set custom config directory
export OPENCODE_CONFIG_DIR=/path/to/config

# Set API keys for providers
export ANTHROPIC_API_KEY=your-key
export ZHIPU_API_KEY=your-key

Configuration Files

OpenCode automatically loads configuration from:

• Global: ~/.config/opencode/
• Project: .opencode/ (in project root)

Agent Configuration

Custom agents go in .opencode/agent/:

---
description: Specialized agent for documentation
mode: subagent
tools:
  read: true
  grep: true
permissions:
  edit: allow
  bash:
    "git*": allow
    "*": ask
---

You are an expert technical documentation writer...

Basic API Usage

Create Session

curl -X POST http://localhost:3000/session \
  -H "Content-Type: application/json" \
  -d '{"title": "My Development Session"}'

Send Message

curl -X POST http://localhost:3000/session/{session_id}/message \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "general",
    "model": {"providerID": "zai-coding-plan", "modelID": "glm-4.6"},
    "parts": [{"type": "text", "text": "Help me understand this codebase"}]
  }'

List Sessions

curl http://localhost:3000/session | jq '.'

Read Files

curl "http://localhost:3000/file/content?path=README.md"

Swarm Health Monitoring

Health Check

# Check multiple servers
for port in 3001 3002 3003; do
    if curl -s "http://localhost:$port/config" > /dev/null; then
        echo "✅ Server on port $port - HEALTHY"
    else
        echo "❌ Server on port $port - DOWN"
    fi
done

Monitor Active Sessions

for port in 3001 3002 3003; do
    if curl -s "http://localhost:$port/config" > /dev/null; then
        sessions=$(curl -s "http://localhost:$port/session" | jq '. | length')
        echo "Port $port: $sessions active sessions"
    fi
done

Production Deployment

Docker Setup

# Dockerfile
FROM node:18-alpine

RUN npm install -g opencode-ai

EXPOSE 3000

CMD ["opencode", "serve", "--port", "3000", "--hostname", "0.0.0.0"]

Docker Compose

# docker-compose.yml
version: '3.8'
services:
  opencode-1:
    build: .
    ports:
      - "3001:3000"
    environment:
      - OPENCODE_LOG_LEVEL=info
    volumes:
      - ./config:/app/.opencode

  opencode-2:
    build: .
    ports:
      - "3002:3000"
    environment:
      - OPENCODE_LOG_LEVEL=info
    volumes:
      - ./config:/app/.opencode

Best Practices

Security

# Use API keys
export OPENCODE_API_KEY=your-secure-key

# Restrict access to localhost
opencode serve --hostname 127.0.0.1

# Use reverse proxy for SSL
# Configure nginx/cloudflare for HTTPS

Performance

# Set appropriate log levels
export OPENCODE_LOG_LEVEL=warn

# Monitor server resources
curl http://localhost:3000/config | jq '.providers'

Port Management

# Find available port
python -c "import socket; s=socket.socket(); s.bind(('', 0)); print(s.getsockname()[1]); s.close()"

# Kill process using port
lsof -ti:3000 | xargs kill -9

Troubleshooting

Common Issues

# Check if port is available
netstat -tulpn | grep :3000

# Kill existing server
pkill -f "opencode serve"

# Test server response
curl -v http://localhost:3000/config

This skill provides everything needed to launch, configure, and manage OpenCode servers from basic single instances to multi-server swarms, all using simple commands and tools that are already available.