Skip to content

CLAUDE.md

Latest commit

 

History

History
42 lines (32 loc) · 2.01 KB

CLAUDE.md

File metadata and controls

42 lines (32 loc) · 2.01 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

For general usage instructions, Docker examples, and environment configuration:

@./README.md

Architecture Overview

Core Components

  • main.py: Entry point that creates and configures the root agent using the Microsoft Agent Framework (MSAF)
  • Agent Configuration: The root agent is configured via environment variables and supports:
    • Dynamic sub-agent loading from URL-based agent cards
    • MCP (Model Context Protocol) toolset integration
    • OpenAI-compatible model integration via LiteLLM proxy

Agent System Design

  • Root Agent: Main agent created with Agent() class from agent_framework
  • Sub-Agents: Remote A2A (Agent-to-Agent) agents loaded from environment configuration
  • Tools: MCP toolsets that provide external functionality via HTTP connections
  • Model Integration: Uses OpenAI-compatible clients supporting various LLM providers via gateway

Configuration Pattern

The application uses a JSON-based configuration system through environment variables:

  • SUB_AGENTS: Maps agent names to configuration objects with URLs
  • AGENT_TOOLS: Maps tool names to configuration objects with connection parameters
  • Both configurations are parsed from JSON and have error handling for invalid JSON

Deployment

  • Multi-stage Docker build using Python 3.14 with uv for dependency management
  • Exposes port 8000 by default
  • Supports multi-platform builds (linux/arm64, linux/amd64)

Dependencies

  • agentic-layer-sdk-msaf: Core MSAF functionality for agent creation
  • agent-framework: Microsoft Agent Framework
  • uvicorn: ASGI server for serving the agent API
  • OpenAI-compatible gateway: Model abstraction layer (e.g. LiteLLM proxy)
  • MCP integration: For external tool connectivity

The codebase follows a template pattern for creating configurable agents that can be extended with sub-agents and tools through environment configuration rather than code changes.