MCP--AI-assistant-

### **Project: AI Assistant with MCP Servers (LangChain + Groq)** **Technologies:** MCP (Model Context Protocol), LangChain, Groq (Llama3-8B), Cursor IDE (MCP Host), Streamlit/Flask, MCP Servers (Playwright, Airbnb, DuckDuckGoSearch) #### **Key Features & Achievements** 1. **MCP-Powered Multi-Source Integration** - Built an AI assistant that dynamically connects to **MCP servers** (Playwright for web automation, Airbnb/DuckDuckGo for real-time data) via a configurable `MCPClient`. - Leveraged **LangChain's agent framework** to orchestrate context-aware interactions between Groq’s LLM (`llama3-8b-8192`) and MCP services. 2. **Conversational Memory & Stateful Dialog** - Implemented persistent conversation history (`memory_enabled=True`) to retain context across user queries, enabling coherent long-term interactions. - Added user commands (`clear`, `exit`) to manage session state dynamically. 3. **Scalable & Fault-Tolerant Design** - Used **asynchronous I/O** (`asyncio`) for non-blocking MCP server calls, ensuring responsiveness. - Deployed error handling and session cleanup to gracefully manage API failures (`client.close_sessions()`). #### **Technical Highlights** - **MCP Protocol**: Customized `MCPAgent` to route queries between LLM and external services (e.g., scrape data via Playwright, fetch listings from Airbnb). - **Groq Optimization**: Utilized Groq’s ultra-low-latency API for real-time inference. - **Config-Driven**: Server endpoints defined in `browser_mcp.json` for modularity. **Impact:** Enabled accurate, multi-source responses (e.g., travel queries combining Airbnb data + DuckDuckGo search) with a single chatbot interface. --- ### **Code Snippet Summary** ```python # 1. Initialize MCP client with servers (Playwright, Airbnb, etc.) client = MCPClient.from_config_file("browser_mcp.json") # 2. Set up Groq's Llama3 LLM llm = ChatGroq(model="llama3-8b-8192") # 3. Create agent with memory agent = MCPAgent(llm=llm, memory_enabled=True, client=client) # 4. Run interactive chat loop response = await agent.run(user_input) # Fetches data from MCP servers + LLM ``` --- ### **Why It Stands Out** - **Beyond RAG**: Unlike static retrieval, your MCP agent **actively pulls live data** from diverse APIs/web sources. - **User-Centric**: Memory and session controls mimic commercial chatbots (e.g., ChatGPT).