nedops/launch-mode-mcp

🚀 Launch Mode MCP Server

Real-Time Code Validation for AI Assistants

Save 97.5% of your AI assistant's context by validating code changes externally. Get instant feedback without consuming precious tokens.

✅ NOW MCP COMPATIBLE - Install directly in Claude Desktop or Claude Code!
📊 - Real test shows 8,000 → 200 tokens per file!

🎯 What It Does

When your AI assistant writes code, it typically uses significant context to:

• Check syntax
• Run tests
• Validate API contracts
• Ensure compatibility

Launch Mode MCP handles all of this externally, giving you:

• ✅ Sub-second validation
• ✅ Real test results
• ✅ Breaking change detection
• ✅ Service dependency awareness

⚡ Quick Start (30 seconds)

For Claude Code (CLI)

# Install the MCP server
claude mcp add launch-mode /path/to/venv/bin/python /path/to/launch_mode_mcp.py

# Or with the provided path:
claude mcp add launch-mode /Users/ned/repos/github/nedops/launch-mode-mcp/venv/bin/python /Users/ned/repos/github/nedops/launch-mode-mcp/launch_mode_mcp.py

For Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "launch-mode": {
      "command": "/path/to/venv/bin/python",
      "args": ["/path/to/launch_mode_mcp.py"]
    }
  }
}

For REST API Server (Legacy)

git clone https://github.com/nedops/launch-mode-mcp
cd launch-mode-mcp
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python mcp_server.py  # Runs on http://localhost:8000

Test It Works

curl http://localhost:8000/health

🤖 Available Tools

Once installed, these MCP tools are available:

• validate_changes - Validate code syntax and run tests

  • Supports Python, JavaScript, TypeScript
  • Runs pytest and jest automatically
  • Detects common issues and anti-patterns

• get_service_info - Get service configuration

  • View dependencies and API endpoints
  • Check service health status

• run_smoke_test - Execute smoke tests

  • Health checks
  • API validation
  • Integration tests

• list_services - List all configured services

📊 Example Usage

Validate Code Changes

# Your AI writes this code
def calculate_total(items):
    return sum(item.price for item in items)

# Launch Mode MCP validates it instantly:
# ✅ Syntax valid
# ✅ 5 tests passed
# ⚠️ Warning: Missing type hints
# 💡 Suggestion: Add error handling for empty list

Check Breaking Changes

# Changing an API endpoint
@app.route('/api/users/<id>', methods=['DELETE'])  # was GET

# Launch Mode MCP warns:
# ❌ Breaking change detected!
# - Method changed from GET to DELETE
# - 3 services depend on this endpoint
# - Suggested: Version the API instead

🛠️ Configuration

Edit services.yaml to add your services:

services:
  - name: your-service
    repository: https://github.com/you/repo
    health_endpoint: /health
    dependencies: [database, cache]
    api_endpoints:
      - GET /api/resource
      - POST /api/resource

📈 Current Status & Roadmap

✅ Skateboard (Current - v1.0)

• ✅ MCP protocol implementation
• ✅ Python/JavaScript validation
• ✅ Real pytest/jest execution
• ✅ Claude Desktop/Code integration
• ✅ Common issue detection
• ✅ API breaking change warnings

🚲 Bicycle (Next - v2.0)

• 🔄 Cross-service validation
• 🔄 Real-time test watching
• 🔄 More language support (Go, Rust)
• 🔄 Caching for faster responses
• 🔄 Integration test coordination

🚗 Car (Future - v3.0)

• 📅 Full context persistence
• 📅 Multi-agent orchestration
• 📅 Autonomous test generation
• 📅 Semantic code understanding

🎯 Why This Matters

Proven Results (September 1, 2025 Test)

Traditional AI coding workflow:

Reading file:     5,000 tokens
Validation:       3,000 tokens  
Total per file:   8,000 tokens (4% of context)
20 files:       160,000 tokens (80% of context!) ⚠️

With Launch Mode MCP:

Reading file:         0 tokens (reads from disk)
Validation:         200 tokens (just results)
Total per file:     200 tokens (0.1% of context)
20 files:         4,000 tokens (2% of context!) ✨

📊 97.5% Context Savings -

🏗️ Architecture

Two Server Implementations

1. launch_mode_mcp.py - Standard MCP Server

   • Uses official MCP Python SDK
   • stdio transport for AI assistants
   • Direct integration with Claude

2. mcp_server.py - REST/WebSocket Server

   • HTTP endpoints for web integration
   • WebSocket for real-time validation
   • Standalone operation on port 8000

Validation Features

• Syntax Checking: Python (compile), JavaScript (bracket matching)
• Test Execution: pytest with JSON reporting, jest support
• Issue Detection: TODOs, console.logs, hardcoded values, wildcard imports
• API Validation: Breaking change detection, contract checking

📊 Metrics

After validation, you get:

• ✅ Pass/fail status
• 📝 List of errors
• ⚠️ Warnings
• 💡 Suggestions
• ⏱️ Execution time
• 🧪 Number of tests run

🤝 Contributing

This is the "skateboard" version - simple but functional. Help us evolve:

1. Use it and report issues
2. Extend it with your validators
3. Share it with your team
4. Contribute improvements

📚 Documentation

• - Connect your AI assistant
• - Detailed API documentation
• - Extend and customize

🐛 Troubleshooting

Server won't start?

# Check port availability
lsof -i :8000

# Check logs
docker-compose logs

Validation timing out?

# Increase timeout
export TEST_TIMEOUT=60

AI not connecting?

# Verify server is running
curl http://localhost:8000/health

📄 License

MIT - Use freely in your projects

🙏 Acknowledgments

Part of the Launch Mode platform - simplifying AI-assisted development by solving the context problem.

Built with frustration from wasted context, for developers who want their AI assistants to actually help code, not just validate it.