Build your own Local MCP Client with LlamaIndex

This project demonstrates how to build a local MCP (Model Context Protocol) client using LlamaIndex. The client connects to a local MCP server (which exposes tools like a SQLite database) and lets you interact with it using natural language and tool-calling agents—all running locally on your machine.

Setup

To sync dependencies, run:

uv sync

---

Usage

• Start the local MCP server (for example, the included SQLite demo server):

uv run server.py --server_type=sse

• Run the client (choose the appropriate client script, e.g. client.py for OpenAI or ollama_client.py for Ollama):

uv run client.py

• Interact with the agent in your terminal. Type your message and the agent will use the available tools to answer your queries.