MCP is just a format/protocol, ContextVM uses Nostr as a transport layer, but where do you get the inference powers of the LLMs in the end?
You plug the MCPs into the LLMs, so you can run inference from whichever source you prefer. It’s like giving the LLM or agent some tools. The tools are described and can be invoked through the MCP. In any agentic framework, such as Claude Code, Cursor, or any other, you can configure which MCPs you’ll provide to the LLM.