Model Context Protocol Demo

# MCP : Tool and Resource ## 🆕 New Features Added: ### 1. **Resource Discovery & Usage** - Added resource listing with `client.list_resources()` - Enhanced LLM parser to understand resource queries - Added resource reading with `client.read_resource(uri)` ### 2. **Enhanced Query Parsing** - **LLM Parser**: Now recognizes resource requests like "Tell me about Apple as a company" - **Rule-based Parser**: Enhanced to distinguish between tool calls and resource reads - **Combined Actions**: Can execute both tool calls and resource reads for queries like "Get Apple stock price and company info" ### 3. **New Demo Commands** - `resources` - Lists available resources - Enhanced help text with resource examples ### 4. **Standalone Resource Demo** - Added `demo_resources()` function for focused resource testing - Demonstrates reading `info://server` and `stock://{ticker}` resources - Shows combined tool + resource usage ### 5. **Smart Query Examples** The client now handles queries like: - **Resource-only**: "Tell me about Apple as a company" → reads `stock://AAPL` - **Tool-only**: "What's Apple's stock price?" → calls `stock_quote` tool - **Combined**: "Get Apple stock price and company info" → both tool + resource ## 🚀 Usage Examples: ```bash # Interactive mode with resource support python mcp_tool_resource.py # Resource-only demonstration python mcp_tool_resource.py # Then choose option 2 ``` ## 🎯 Try These Queries: - "Tell me about Microsoft as a company" - "What is Tesla?" - "Give me server information" - "Get Apple stock price and company info" - "I want both Tesla's price and company details" The enhanced client now provides a complete MCP experience with both tools and resources, powered by intelligent LLM parsing that can distinguish between different types of requests! ## ✅ **The resources ARE working correctly!** Here's what we can see: 1. **Only `info://server` shows up in `list_resources()`** - This is expected because it's a static resource 2. **The templated `stock://{ticker}` resource pattern doesn't appear in the list** - This is normal MCP behavior for dynamic resources 3. **BUT the resources actually work!** - ✅ `stock://AAPL` successfully returns: "Apple Inc. (AAPL) - A publicly traded company" - ✅ `info://server` successfully returns the server description ## 🔧 **The Fix** The issue is in the enhanced MCP client - it's looking for `stock://{ticker}` in the resource list, but dynamic resources don't show up there. We need to update the client to: 1. **Hard-code knowledge of the dynamic resource patterns** since they don't appear in discovery 2. **Still try to read them even if they're not listed**Now your enhanced MCP client will: 1. ✅ **Discover static resources** from the server (like `info://server`) 2. ✅ **Add knowledge of dynamic resources** that don't appear in discovery (like `stock://{ticker}`) 3. ✅ **Successfully read both types** when requested ## 🎯 **Test it now!** Try these queries in your enhanced client: ```bash python mcp_client_llm_resource.py ``` Then test: - `"Tell me about Apple as a company"` → should read `stock://AAPL` - `"Give me server information"` → should read `info://server` - `"Get Apple stock price and company info"` → should do both tool call + resource read The resources are working perfectly - it was just a discovery issue! Dynamic/templated resources in MCP don't show up in `list_resources()` but are still accessible via `read_resource()`. # key conceptual differences between **Tools** and **Resources** in MCP: ## 🔧 **Tools** - Actions & Operations **Tools are functions that DO something** - **Purpose**: Perform actions, calculations, or operations - **Nature**: Dynamic, stateful, can change things - **Input**: Accept parameters to customize behavior - **Output**: Return results based on computation/action - **Examples**: - `calculator(operation="add", num1=15, num2=27)` → performs calculation - `stock_quote(ticker="AAPL")` → fetches live data from APIs - `send_email(to="user@example.com", subject="Hello")` → performs action ```python @mcp.tool() def calculator(operation: str, num1: float, num2: float) -> dict: # DOES something - performs calculation return {"result": num1 + num2} ``` ## 📚 **Resources** - Information & Content **Resources are content that IS something** - **Purpose**: Provide static/semi-static information or content - **Nature**: Informational, read-only, descriptive - **Input**: URI-based addressing (like web URLs) - **Output**: Return content/information about something - **Examples**: - `info://server` → returns server description - `stock://AAPL` → returns company information about Apple - `file://report.pdf` → returns document content ```python @mcp.resource("stock://{ticker}") def stock_info(ticker: str) -> str: # IS something - returns information return f"Apple Inc. (AAPL) - A publicly traded company" ``` ## 🆚 **Key Differences** | Aspect | Tools | Resources | |--------|-------|-----------| | **Verb vs Noun** | "Calculate this" | "What is this?" | | **Action vs Info** | Performs operations | Provides information | | **Dynamic** | Can change state/external systems | Read-only content | | **Parameters** | Function parameters | URI-based addressing | | **Use Cases** | APIs, calculations, actions | Documentation, descriptions, content | ## 🎯 **Real-World Analogy** Think of it like a **library**: - **Tools** = **Librarian** 📖 - "Please find me books about Python" (search action) - "Calculate my late fees" (computation) - "Check out these books" (state-changing action) - **Resources** = **Books/Content** 📚 - "Python Programming Guide" (static content) - "Library Rules and Policies" (informational content) - "Map of the Library" (descriptive content) ## 🔄 **When to Use Each** **Use Tools when you want to:** - Perform calculations - Fetch live/dynamic data - Change system state - Execute operations - Process inputs **Use Resources when you want to:** - Provide static information - Share documentation - Describe entities/concepts - Offer reference content - Give context about something ## 💡 **Your MCP Server Example** ```python # TOOL - Does something (fetches live stock price) @mcp.tool() def stock_quote(ticker: str) -> dict: # Actively fetches current price from Yahoo Finance API return {"current_price": 150.25, "volume": 1000000} # RESOURCE - Is something (provides company info) @mcp.resource("stock://{ticker}") def stock_info(ticker: str) -> str: # Returns descriptive information about the company return "Apple Inc. (AAPL) - A publicly traded company" ``` This is why your enhanced client can do **both** for queries like *"Get Apple stock price and company info"* - it calls the tool for live data and reads the resource for descriptive information!