MathAgent

A minimal math agent that delegates all calculations to an MCP (Model Context Protocol) server. The agent uses MCP tools for every operation—no mental math—so you get consistent, inspectable results.

What’s inside

• Agent: built with pydantic-ai, connected via MCP stdio
• MCP Server: powered by fastmcp, exposing math tools

Available MCP tools

• add
• subtract
• multiply
• divide
• power
• log

Coming soon:

• modulus

Requirements

• Python 3.13+
• uv (Python package/dependency manager)
• A Groq API key (for the LLM model)
• A Logfire token (for instrumentation)

Setup

1. Clone the repository
2. Create your environment file
   • Set these keys:
     • groq_key — your Groq API key
     • logfire_key — your Logfire token

Run the agent

Run with uv so dependencies are resolved and the MCP server is launched as a toolset:

uv run agent.py

Notes:

• The agent is configured to use the Groq model qwen/qwen3-32b (see agent.py). You can swap models as needed.
• The agent will reject non-math questions by design.
• All calculations route through the MCP tools defined in mcp-server.py.

Contributing

Contributions are welcome—feel free to open issues or submit pull requests.