10 March 2024
I have been reading AI related stuff on X and trying them out on my own. Here are the few of my bookmarks and why I think it’s worth taking a check.
CopilotKit team created MCP Client which use CopilotKit on FE and LangGraph.
I have been figuring out how MCP Client will work on the web. I have done some testing on my own and I did not really get it. But with CopilotKit implementation I understand it fully. With this, I think we can easily create agents that are able to answer questions. The following image is how I think about using MCP. It’s just one case.
Alex Albert tweeted claude + mcp client + mcp servers = agent
I think it makes sense. We might probably be able to build something like https://manus.im/app like AI Agent with different LLM agent architectures and workflow patterns. And MCP will make it easier to access tools use.
I saw a few tweets about how they are able to get sandbox runtime code. Peak from Manus AI replied to him with a few clarifications. They don’t use MCP as when they started, there is no MCP.
https://manus.im/share/lLR5uWIR5Im3k9FCktVu0k?replay=1
https://gist.github.com/jlia0/db0a9695b3ca7609c9b1a08dcbf872c9