How nshkrdotcom/agent_experiments works
Project type and purpose
• This is a collection of AI-agent experiments and tooling (“agent_experiments”) written in Python. It includes:
• agent2 and multi_tool_agent packages (core agent implementations)
• an MCP client (“mcp/mcp-client”) with command-line interface, engine, configuration, and services layers
• a sample microservice (“mcp/weather”) demonstrating an external tool the agent can call
• The purpose is to explore multi-agent patterns, external service adapters, and configurable workflows for AI agents.
File structure
• Top level:
– agent2/agent2.py (a standalone agent prototype)
– multi_tool_agent/agent.py (another standalone agent)
– mcp/
· mcp-client/ (a reusable client library & CLI for MCP-style agents)
· weather/ (a simple HTTP-style service module the client can query)
• mcp-client sub-structure follows a modular layout:
– cli.py (entrypoint for user commands)
– engine.py (orchestrator that drives agent workflows)
– services.py (abstraction layer for calling out to external systems or microservices)
– config.py + workflows.json + mcp_servers.json (configuration for endpoints & workflows)
– app_logger.py (logging framework)
• mcp/weather contains a main.py and weather.py, showing how an external service can be spun up or invoked.
README
• Confirms this is an experimental suite for “MCP, Google ADK, and related AI agent experiments.”
Detail
• Break out mcp-client into its submodules (cli, engine, services, config, logger)
• Sample workflows.json driving the engine
• Weather service as an example microservice block