Agents

A minimal educational implementation of LLM agents using the Claude API.

Note: This is NOT an SDK, but a reference implementation of key concepts

Overview & Core Components

This repo demonstrates how to build effective agents with the Claude API. It shows how sophisticated AI behaviors can emerge from a simple foundation: LLMs using tools in a loop. This implementation is not prescriptive - the core logic is <300 lines of code and deliberately lacks production features. Feel free to translate these patterns to your language and production stack (Claude Code can help!)

It contains three components:

• agent.py: Manages Claude API interactions and tool execution
• tools/: Tool implementations (both native and MCP tools)
• utils/: Utilities for message history and MCP server connections

Usage

from agents.agent import Agent
from agents.tools.think import ThinkTool

# Create an agent with both local tools and MCP server tools
agent = Agent(
    name="MyAgent",
    system="You are a helpful assistant.",
    tools=[ThinkTool()],  # Local tools
    mcp_servers=[
        {
            "type": "stdio",
            "command": "python",
            "args": ["-m", "mcp_server"],
        },
    ]
)

# Run the agent
response = agent.run("What should I consider when buying a new laptop?")

From this foundation, you can add domain-specific tools, optimize performance, or implement custom response handling. We remain deliberately unopinionated - this backbone simply gets you started with fundamentals.

Requirements

• Python 3.8+
• Claude API key (set as ANTHROPIC_API_KEY environment variable)
• anthropic Python library
• mcp Python library