MCP + Ollama Local Tool Calling Example

What is MCP + Ollama Local Tool Calling Example?

It is a demonstration project that shows how a local AI agent can understand user queries and automatically call Python functions using the Model Context Protocol (MCP), a local LLM via Ollama, and a Python MCP client and server.

How to use MCP + Ollama Local Tool Calling Example?

Install dependencies with pip install "mcp[cli] @ git+https://github.com/awslabs/mcp.git" openai==0.28 httpx, ensure Ollama is installed and running, pull a tool-calling‑capable model (e.g., ollama run llama3), run the MCP server (python math_server.py), then start the client with python ollama_client.py math_server.py. Finally, type queries like “What is 5 + 8?” and the system will respond with the computed result.

Key features of MCP + Ollama Local Tool Calling Example

• Uses MCP to describe tools for the LLM
• Runs a local LLM (e.g., Llama3) via Ollama
• Python functions (add, multiply) become callable tools
• Fully autonomous – no manual tool selection
• Everything runs offline and locally

Use cases of MCP + Ollama Local Tool Calling Example

• Building a local AI developer assistant that can perform calculations
• Creating secure, offline‑only AI workflows for sensitive data
• Prototyping autonomous agents that reason about and invoke functions

FAQ from MCP + Ollama Local Tool Calling Example

What does this project demonstrate?

It demonstrates how an LLM can automatically decide which Python function to call based on a user’s intent, using MCP to describe available tools and Ollama to run the model locally.

What are the runtime requirements?

You need Python, the mcp[cli], openai==0.28, and httpx packages, plus Ollama installed and running with a model that supports tool calling (e.g., Llama3).

Does the system require an internet connection?

No – everything runs locally. The LLM is served by Ollama on your machine, and the MCP components communicate over local transports.

Which tools are exposed by the MCP server?

The example server (math_server.py) exposes two tools: add(a: int, b: int) -> int and multiply(a: int, b: int) -> int.

Can I use a different LLM model?

Yes, as long as the model supports tool‑calling. The README suggests Llama3 via the ollama run llama3 command; you can replace it with any compatible Ollama model.