mcp (model context protocol) is a way to expose any kind of functionality to the world of ai agents
we start with a llm (large language model), then move to an agent (build on top of the AI SDK) and providing it with a tool to get the weather for a given location
we build our own mcp server and let our agent use it
we take a look at agent frameworks like mastra to build ai apps in cursor and see how we can use the mcp server from mastra to get help during development
we explore what lerobot is and how we can control a robot with pi0 (live demo using SO-100 + leader arm)
we create a tool that takes a prompt (20 tokens max) to describe the action that the robot arm should be doing. this tool is used by our agent to execute a list of steps to reach a certain goal, for example:
"please pick up the pen and throw it away"
"can you please clean up this mess?"
"where are my keys?"
we see how ai can interact with the real world, using open-source hardware & software