I get the client part, great job!

However, the agent is really just a wrapper of Langchain AgentExecutor. This doesn't seem like something someone would want to put into production.

Valid feedback, we did not put as much work on the agent as we did for the client and MCP side yet, but we plan to write a cleaner more composable MCPAgent as well. Actually I would be super happy to hear what you consider a standard here, something like https://openai.github.io/openai-agents-python/ ?