MCP TypeScript Demo Server

import 'dotenv/config'; import { Client } from '@modelcontextprotocol/sdk/client/index.js'; import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js'; import readline from 'readline'; import Anthropic from '@anthropic-ai/sdk'; import { getToolSelectionPrompt } from './index.js'; const ANTHROPIC_API_KEY = process.env.ANTHROPIC_API_KEY || ''; async function askLLM(userQuery: string, tools: any[]): Promise<{ tool: string; arguments: any }> { const anthropic = new Anthropic({ apiKey: ANTHROPIC_API_KEY }); // Create a detailed tool list with exact argument specifications const toolList = tools.map((t: any) => { const args = t.inputSchema?.properties ? Object.entries(t.inputSchema.properties).map(([key, prop]: [string, any]) => `${key} (${prop.type}${prop.description ? `: ${prop.description}` : ''}${t.inputSchema.required && t.inputSchema.required.includes(key) ? ', required' : ', optional'})` ).join(', ') : 'no arguments'; return `- ${t.name}: ${t.description || ''} - Arguments: { ${args} }`; }).join('\n'); // Use the imported prompt template const prompt = getToolSelectionPrompt(toolList, userQuery); const message = await anthropic.messages.create({ model: 'claude-3-7-sonnet-20250219', max_tokens: 1000, messages: [ { role: 'user', content: prompt } ] }); const text = message.content[0].type === 'text' ? message.content[0].text : ''; try { return JSON.parse(text); } catch { throw new Error('Failed to parse LLM output: ' + text); } } async function main() { // Set up the MCP client const serverUrl = 'http://localhost:3000/mcp'; const transport = new StreamableHTTPClientTransport(new URL(serverUrl)); const client = new Client({ name: 'example-client', version: '1.0.0' }); // Connect to the server await client.connect(transport); console.log('Connected to MCP server.'); // List available tools const toolsRaw = await client.listTools(); console.log('Raw tools response:', JSON.stringify(toolsRaw, null, 2)); // Handle different possible formats of tools response let toolsArr: any[] = []; if (toolsRaw && typeof toolsRaw === 'object' && 'tools' in toolsRaw) { // Handle nested structure: { tools: [...] } toolsArr = Array.isArray(toolsRaw.tools) ? toolsRaw.tools : []; } else if (Array.isArray(toolsRaw)) { // Handle direct array toolsArr = toolsRaw; } else if (toolsRaw && typeof toolsRaw === 'object') { // Handle object with tool properties toolsArr = Object.values(toolsRaw); } else { console.error('Unexpected tools format:', typeof toolsRaw); return; } // Filter out any undefined or null tools and ensure proper structure toolsArr = toolsArr.filter(tool => tool && typeof tool === 'object' && tool.name); console.log('Available tools:'); toolsArr.forEach((tool, index) => { console.log(`${index + 1}. ${tool.name}: ${tool.description || 'No description'}`); }); if (toolsArr.length === 0) { console.error('No tools available. Please check if the server is running and has registered tools.'); return; } // Set up readline for chatbot loop const rl = readline.createInterface({ input: process.stdin, output: process.stdout }); async function promptUser(query: string): Promise<string> { return new Promise((resolve) => rl.question(query, resolve)); } while (true) { const userInput = await promptUser('\nAsk me anything (or type "exit" to quit): '); const trimmedInput = userInput.trim().toLowerCase(); if (trimmedInput === 'exit' || trimmedInput === 'quit') break; try { const { tool, arguments: toolArgs } = await askLLM(userInput, toolsArr); if (!tool || !toolsArr.some(t => t.name === tool)) { console.log('No suitable tool found for your query.'); continue; } console.log(`\nLLM selected tool: ${tool}`); console.log('With arguments:', toolArgs); const result = await client.callTool({ name: tool, arguments: toolArgs }); console.log('\nTool result:'); console.dir(result, { depth: null }); } catch (err) { console.error('Error:', err); } } rl.close(); console.log('Client finished.'); process.exit(0); } main().catch((err) => { console.error('Client error:', err); process.exit(1); });