AI MCP Gateway

#!/usr/bin/env node /** * MCP CLI - Command Line Interface for AI MCP Gateway * * Usage: * mcp chat "your message" - Send single message * mcp chat - Interactive chat mode * mcp code path/to/file.js -p "..." - Analyze code file * mcp code - < file.js - Analyze from stdin * mcp diff path/to/file.js -p "..." - Get diff patch * * Environment Variables: * MCP_ENDPOINT - API endpoint (default: http://localhost:3000) * MCP_API_KEY - API authentication key */ import { Command } from 'commander'; import chalk from 'chalk'; import { chatCommand } from './commands/chat.js'; import { codeCommand } from './commands/code.js'; import { diffCommand } from './commands/diff.js'; import { analyzeCommand } from './commands/analyze.js'; import { createProjectCommand } from './commands/create-project.js'; import { claudeCommand } from './commands/claude.js'; import { summarizeProject } from './commands/summarize.js'; import { debugCommand } from './commands/debug.js'; import { mcpServeCommand } from './commands/mcp-serve.js'; const program = new Command(); program .name('mcp') .description('CLI tool for AI MCP Gateway') .version('0.1.0'); // Global options program .option('-e, --endpoint <url>', 'MCP server endpoint URL') .option('-k, --api-key <key>', 'API authentication key') .option('-u, --username <username>', 'Admin username for authentication') .option('-p, --password <password>', 'Admin password for authentication'); // Chat command program .command('chat [message]') .description('Chat with AI (interactive mode if no message provided)') .option('-i, --interactive', 'Force interactive mode') .option('--use-claude-code', 'Use Claude Code instead of multi-layer API') .option('-a, --agent', 'Enable agent mode (Claude Code / Copilot-like)') .option('--auto-execute', 'Auto-execute actions without confirmation') .option('-b, --budget <number>', 'Budget for API calls', parseFloat) .action(async (message: string | undefined, options: Record<string, unknown>) => { const globalOpts = program.opts(); await chatCommand(message, { endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, interactive: options.interactive as boolean | undefined, useClaudeCode: options.useClaudeCode as boolean | undefined, agent: options.agent as boolean | undefined, autoExecute: options.autoExecute as boolean | undefined, budget: options.budget as number | undefined, }); }); // Code command program .command('code <file>') .description('Send code file for AI analysis/review or create new code (GitHub Copilot-style)') .option('-p, --prompt <text>', 'Custom prompt for AI') .option('--stdin', 'Read from stdin instead of file') .option('--no-context', 'Disable context analysis (faster but less accurate)') .option('--no-related', 'Skip related files analysis') .option('-c, --create', 'Create new code instead of analyzing existing file') .option('-o, --output <file>', 'Output file for generated code') .option('--apply', 'Apply generated changes to disk') .option('--use-claude-code', 'Use Claude Code instead of multi-layer API') .action(async (arg: string, options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await codeCommand(arg, { prompt: options.prompt as string | undefined, endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, stdin: options.stdin as boolean | undefined, context: options.context as boolean | undefined, related: options.related as boolean | undefined, create: options.create as boolean | undefined, output: options.output as string | undefined, apply: options.apply as boolean | undefined, useClaudeCode: options.useClaudeCode as boolean | undefined, }); }); // Diff command program .command('diff <file>') .description('Request unified diff patch from AI') .option('-p, --prompt <text>', 'Custom prompt for changes') .option('--apply', 'Automatically apply the patch (not implemented)') .option('--use-claude-code', 'Use Claude Code instead of multi-layer API') .action(async (arg: string, options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await diffCommand(arg, { prompt: options.prompt as string | undefined, endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, apply: options.apply as boolean | undefined, useClaudeCode: options.useClaudeCode as boolean | undefined, }); }); // Analyze command - Multi-file analysis program .command('analyze <pattern>') .description('Analyze multiple files (e.g., "src/**/*.ts")') .option('-p, --prompt <text>', 'Custom analysis prompt') .option('-m, --max-files <number>', 'Maximum files to analyze (default: 10)', '10') .option('-r, --recursive', 'Include subdirectories') .option('--use-claude-code', 'Use Claude Code instead of multi-layer API') .action(async (pattern: string, options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await analyzeCommand(pattern, { prompt: options.prompt as string | undefined, endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, maxFiles: parseInt(options.maxFiles as string), recursive: options.recursive as boolean | undefined, useClaudeCode: options.useClaudeCode as boolean | undefined, }); }); // Create Project command - AI-powered project scaffolding program .command('create-project [description]') .description('Create a complete project with AI (auto-structure, tests, CI/CD). If no description provided, looks for mcp-instruction.md or mcp-instructor.md') .option('-o, --output <dir>', 'Output directory (default: auto-generated name)') .option('--no-install', 'Skip npm install') .option('--no-test', 'Skip initial test run') .option('-v, --verbose', 'Verbose output') .option('-b, --budget <amount>', 'Budget for AI generation (USD)', parseFloat) .option('-l, --max-layer <layer>', 'Maximum layer to use (L0/L1/L2/L3)') .option('--enable-test', 'Enable testing during generation') .option('--enable-debug', 'Enable debug mode') .option('--use-claude-code', 'Use Claude Code engine instead of multi-layer API') .action(async (description: string | undefined, options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await createProjectCommand(description, { endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, output: options.output as string | undefined, budget: options.budget as number | undefined, maxLayer: options.maxLayer as string | undefined, noTests: options.noTest as boolean | undefined, debug: options.enableDebug as boolean | undefined, useClaudeCode: options.useClaudeCode as boolean | undefined, }); }); // Claude command - Launch Claude Code program .command('claude') .description('Launch Claude Code in current or specified directory') .option('--cwd <dir>', 'Working directory for Claude Code') .allowUnknownOption() // Allow forwarding args to claude .action(async (options: Record<string, unknown>, cmd: Command) => { // Extract args after 'claude' command to forward to Claude Code const rawArgs = process.argv.slice(2); // All args after 'node script.js' const claudeIndex = rawArgs.findIndex(arg => arg === 'claude'); const forwardArgs = claudeIndex >= 0 ? rawArgs.slice(claudeIndex + 1).filter(arg => !arg.startsWith('--cwd')) : []; await claudeCommand({ cwd: options.cwd as string | undefined, forwardArgs: forwardArgs.length > 0 ? forwardArgs : undefined, }); }); // Summarize command - Generate comprehensive project summary program .command('summarize') .description('Generate comprehensive project summary by analyzing all files and history') .option('-o, --output <file>', 'Output file for summary (default: PROJECT-SUMMARY-YYYY-MM-DD.md)') .option('-b, --budget <amount>', 'Budget for AI analysis (USD)', parseFloat) .option('-m, --model <model>', 'Specific model to use for analysis') .option('-v, --verbose', 'Show verbose analysis process') .action(async (options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await summarizeProject({ output: options.output as string | undefined, budget: options.budget as number | undefined, model: options.model as string | undefined, verbose: options.verbose as boolean | undefined, endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, }); }); // Debug command - Capture and analyze CLI output program .command('debug <action>') .description('Capture and analyze CLI output for debugging (capture|read|clear|analyze)') .option('-o, --output <file>', 'Output file for captured data') .option('-a, --append', 'Append to existing file instead of overwriting') .option('-f, --format <type>', 'Output format: json, text, markdown (default: text)', 'text') .option('-t, --include-timestamp', 'Include timestamp in output') .option('-l, --max-lines <number>', 'Maximum lines to display when reading', parseInt) .action(async (action: string, options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await debugCommand(action, { output: options.output as string | undefined, append: options.append as boolean | undefined, format: options.format as 'json' | 'text' | 'markdown' | undefined, includeTimestamp: options.includeTimestamp as boolean | undefined, maxLines: options.maxLines as number | undefined, endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, username: globalOpts.username as string | undefined, password: globalOpts.password as string | undefined, }); }); // MCP Server command - Start MCP server for external clients program .command('mcp-serve') .description('Start MCP server for Claude Desktop, VSCode, or other MCP clients') .option('-t, --transport <type>', 'Transport type: stdio, websocket (default: stdio)', 'stdio') .option('-p, --port <number>', 'WebSocket port (default: 3001)', parseInt) .option('-l, --log-level <level>', 'Log level: debug, info, warn, error (default: info)', 'info') .action(async (options: Record<string, unknown>) => { const globalOpts = program.opts() as Record<string, unknown>; await mcpServeCommand({ transport: (options.transport as 'stdio' | 'websocket') || 'stdio', port: (options.port as number) || 3001, logLevel: (options.logLevel as 'debug' | 'info' | 'warn' | 'error') || 'info', endpoint: globalOpts.endpoint as string | undefined, apiKey: globalOpts.apiKey as string | undefined, }); }); // Help command program .command('help') .description('Show detailed help and examples') .action(() => { printDetailedHelp(); }); // Error handling program.exitOverride((err) => { if (err.code === 'commander.help') { process.exit(0); } if (err.code === 'commander.version') { process.exit(0); } console.error(chalk.red(`\n❌ Error: ${err.message}\n`)); process.exit(1); }); // Parse arguments program.parse(process.argv); // Show help if no command provided if (!process.argv.slice(2).length) { program.outputHelp(); } /** * Print detailed help with examples */ function printDetailedHelp(): void { console.log(chalk.cyan('\n╔══════════════════════════════════════════════════════════╗')); console.log(chalk.cyan('║') + chalk.bold(' MCP CLI - AI Coding Assistant ').padEnd(58) + chalk.cyan('║')); console.log(chalk.cyan('╚══════════════════════════════════════════════════════════╝\n')); console.log(chalk.yellow('📖 COMMANDS:\n')); console.log(chalk.white(' mcp chat [message]')); console.log(chalk.dim(' Chat with AI. If no message, starts interactive mode.\n')); console.log(chalk.white(' mcp code <file> [-p "prompt"]')); console.log(chalk.dim(' Send file for code review or analysis.')); console.log(chalk.dim(' Use "-" as filename to read from stdin.\n')); console.log(chalk.white(' mcp analyze <pattern> [-m max]')); console.log(chalk.dim(' Analyze multiple files matching a pattern.\n')); console.log(chalk.white(' mcp create-project <description>')); console.log(chalk.dim(' Create complete project with AI (structure, tests, CI/CD).\n')); console.log(chalk.white(' mcp summarize')); console.log(chalk.dim(' Generate comprehensive project summary analyzing all files and history.\n')); console.log(chalk.white(' mcp diff <file> [-p "prompt"]')); console.log(chalk.dim(' Request AI to generate unified diff patch.\n')); console.log(chalk.white(' mcp debug <action>')); console.log(chalk.dim(' Capture and analyze CLI output for debugging.')); console.log(chalk.dim(' Actions: capture, read, clear, analyze\n')); console.log(chalk.white(' mcp mcp-serve')); console.log(chalk.dim(' Start MCP server for Claude Desktop, VSCode, or other MCP clients.')); console.log(chalk.dim(' Exposes AI routing, network tools, and ops tools via JSON-RPC.\n')); console.log(chalk.yellow('🔧 OPTIONS:\n')); console.log(chalk.white(' -e, --endpoint <url>')); console.log(chalk.dim(' Override MCP_ENDPOINT env variable\n')); console.log(chalk.white(' -k, --api-key <key>')); console.log(chalk.dim(' Override MCP_API_KEY env variable\n')); console.log(chalk.yellow('💡 EXAMPLES:\n')); console.log(chalk.green(' # Interactive chat')); console.log(chalk.dim(' $ mcp chat\n')); console.log(chalk.green(' # Single message')); console.log(chalk.dim(' $ mcp chat "explain async/await in JavaScript"\n')); console.log(chalk.green(' # Code review')); console.log(chalk.dim(' $ mcp code src/index.ts -p "find bugs and suggest improvements"\n')); console.log(chalk.green(' # Analyze multiple files')); console.log(chalk.dim(' $ mcp analyze "src/**/*.ts" -m 5\n')); console.log(chalk.green(' # Create a new project')); console.log(chalk.dim(' $ mcp create-project "React dashboard with authentication"\n')); console.log(chalk.green(' # Generate project summary')); console.log(chalk.dim(' $ mcp summarize -v\n')); console.log(chalk.green(' # Analyze from stdin')); console.log(chalk.dim(' $ cat myfile.js | mcp code -\n')); console.log(chalk.green(' # Get diff patch')); console.log(chalk.dim(' $ mcp diff src/util.py -p "optimize this function"\n')); console.log(chalk.green(' # Custom endpoint')); console.log(chalk.dim(' $ mcp --endpoint https://my-server.com chat "hello"\n')); console.log(chalk.green(' # Start MCP server (for Claude Desktop config)')); console.log(chalk.dim(' $ mcp mcp-serve --transport stdio\n')); console.log(chalk.yellow('🌍 ENVIRONMENT VARIABLES:\n')); console.log(chalk.white(' MCP_ENDPOINT')); console.log(chalk.dim(' API endpoint URL (default: http://localhost:3000)\n')); console.log(chalk.white(' MCP_API_KEY')); console.log(chalk.dim(' API authentication key (optional)\n')); console.log(chalk.yellow('📚 DOCUMENTATION:\n')); console.log(chalk.dim(' https://github.com/yourusername/ai-mcp-gateway\n')); }