FastMCP Demo

# What is an MCP Server Used For? ## The Big Picture An **MCP (Model Context Protocol) Server** is a bridge between AI assistants (like Claude, ChatGPT, etc.) and the real world. It allows AI models to: 1. **Access external data** (files, databases, APIs) 2. **Perform actions** (send emails, create files, make API calls) 3. **Use predefined prompts** (templates for common tasks) Think of it as giving AI assistants "hands" and "eyes" to interact with your tools and data. ## Real-World Use Cases ### 1. **File System Access** Instead of manually copying files, an AI can: - Read files from your computer - Write new files - Search through directories - Organize files by type/date **Example**: "Find all my PDF files from last month and create a summary document" ### 2. **Database Queries** An AI can query your databases without you writing SQL: - Get customer information - Analyze sales data - Generate reports **Example**: "Show me all customers who purchased in the last 30 days" ### 3. **API Integration** Connect AI to external services: - Send emails via Gmail/SendGrid - Create calendar events - Post to social media - Check weather, stocks, news **Example**: "Send an email to my team about tomorrow's meeting" ### 4. **Code Operations** AI can interact with your codebase: - Read code files - Search for functions - Create new files - Run tests **Example**: "Find all functions that use the database connection and show me their error handling" ### 5. **Web Scraping & Research** AI can fetch information from the web: - Get current information - Research topics - Compare prices - Check availability **Example**: "What are the current prices for flights from NYC to London?" ## How It Works ``` ┌─────────────┐ ┌──────────────┐ ┌─────────────┐ │ AI Model │ ◄─────► │ MCP Server │ ◄─────► │ Your Tools │ │ (Claude) │ │ (This Demo) │ │ (Files/APIs)│ └─────────────┘ └──────────────┘ └─────────────┘ User Bridge Real World ``` 1. **User asks AI**: "What files do I have in my Documents folder?" 2. **AI requests tool**: AI calls the MCP server's "list_files" tool 3. **MCP Server executes**: Server reads your file system 4. **Server returns data**: Sends file list back to AI 5. **AI responds to user**: "You have 15 files in Documents..." ## What Our Demo Server Does Our demo server shows three key MCP concepts: ### 1. **Tools** (Actions AI can take) ```typescript // AI can call these functions: - hello(name) → Greets a user - calculate(operation, a, b) → Does math ``` **Real-world equivalent**: - `send_email(to, subject, body)` → Sends an email - `create_file(path, content)` → Creates a file - `query_database(sql)` → Runs a database query ### 2. **Resources** (Data AI can read) ```typescript // AI can read these: - demo://example → Example text - demo://config → Server configuration ``` **Real-world equivalent**: - `file:///Users/you/document.txt` → Read a file - `database://customers` → Query a database table - `api://weather/current` → Get weather data ### 3. **Prompts** (Predefined templates) ```typescript // AI can use these prompt templates: - greet_user(name) → Template for greeting - explain_mcp() → Template for explaining MCP ``` **Real-world equivalent**: - `code_review_prompt(file)` → Template for code review - `email_template(type)` → Email templates - `documentation_prompt(api)` → API documentation template ## Why Use MCP Instead of Direct Integration? ### ✅ **Security** - AI doesn't have direct access to your systems - You control what tools/resources are available - Can add authentication and permissions ### ✅ **Standardization** - One protocol works with all MCP-compatible AI assistants - Write once, use with Claude, ChatGPT, etc. - No need to rebuild for each AI platform ### ✅ **Separation of Concerns** - Your business logic stays in the server - AI just calls tools, doesn't need to know implementation - Easy to update/change without affecting AI ### ✅ **Composability** - Can combine multiple MCP servers - Mix and match capabilities - Build complex workflows ## Example: Building a Personal Assistant MCP Server Here's what a real MCP server might look like: ```typescript // Tools - read_calendar(date) → Get calendar events - create_meeting(title, time, attendees) → Schedule meeting - send_slack_message(channel, message) → Send Slack message - get_weather(location) → Get weather forecast - search_files(query) → Search your files // Resources - file:///notes/* → All your notes - calendar://today → Today's calendar - contacts://all → Your contacts // Prompts - meeting_summary(meeting_id) → Generate meeting summary - email_draft(type, context) → Draft an email ``` Now an AI assistant could: - "Schedule a meeting tomorrow at 2pm with John about the project" - "What's the weather like in San Francisco?" - "Find my notes about the Q4 planning meeting" - "Draft an email to my team about the new feature launch" ## The Bottom Line **MCP Servers turn AI assistants from "chatbots" into "assistants that can actually do things"** Without MCP: AI can only talk and think With MCP: AI can read your files, send emails, query databases, control your tools It's like giving AI a remote control to your digital life (with your permission and control). ## Next Steps To understand MCP better: 1. **Read the code** in `src/index.ts` - see how tools/resources/prompts are defined 2. **Try connecting** a real MCP client (like Claude Desktop) to this server 3. **Build your own tool** - add a tool that does something you need 4. **Explore real examples** - Check out the [MCP Server Examples](https://github.com/modelcontextprotocol/servers)