🧰 MCP Chat – Multi-Server Tool-Enabled Chat App

A Streamlit-based chat application powered by LangChain, OpenAI, and Model Context Protocol (MCP). This project demonstrates how a single LLM can seamlessly interact with multiple MCP servers—both local and remote—to reason, call tools, and return concise final answers.

✨ Built and tested with FastMCP v2, including validation via MCP Inspector.

---

✨ Features

• 💬 Interactive Chat UI using Streamlit

• 🔌 Multi-Server MCP Integration

  • Remote FastMCP cloud server
  • Local Python-based MCP server
  • Node.js MCP server

• 🧠 Tool-aware LLM reasoning with automatic tool selection

• 🧾 Clean chat history (no tool noise or intermediate steps shown)

• 🖥️ Terminal-based MCP chat also implemented (outside the Streamlit UI)

• ☁️ FastMCP Cloud Server tested via MCP Inspector

• 🔄 Async-safe tool execution with proper message ordering

---

🏗️ Architecture Overview

Streamlit UI
    |
    v
LangChain (ChatOpenAI)
    |
    v
MultiServerMCPClient
    |
    +── Remote FastMCP Server (HTTP)
    +── Local Python MCP Server (stdio)
    +── Node.js MCP Server (stdio)

The LLM:

1. Receives user input
2. Decides whether tools are required
3. Executes tools via MCP servers
4. Returns a final concise answer only

---

🔌 Configured MCP Servers

1. Expense Tracker (Remote – FastMCP Cloud)

• Transport: streamable_http
• Deployed on: FastMCP Cloud
• Tested with: MCP Inspector

https://expense-tracker-app.fastmcp.app/mcp

---

2. Manim MCP Server (Local – Python)

• Transport: stdio
• Purpose: Manim-based computation / rendering
• Runs locally using a Python executable

---

3. Sequential Thinking Server (Local – Node.js)

• Transport: stdio
• Powered by: @modelcontextprotocol/server-sequential-thinking
• Used for: Step-by-step structured reasoning

---

⚙️ Tech Stack

• Python 3.12
• Streamlit
• LangChain
• OpenAI (ChatOpenAI)
• FastMCP v2
• Model Context Protocol (MCP)
• Node.js (for MCP servers)
• dotenv

---

📌 Notes

• The system prompt enforces concise final answers
• Tool execution status is never narrated to the user
• Easily extensible—just add more MCP servers to the config!

---

🙌 Acknowledgements

• LangChain
• Model Context Protocol
• FastMCP
• Streamlit
• OpenAI

---