I apologize for this extremely dumb question, but how is this a "server"? As far as I'm aware Blender is a local app. It can run without an internet connection. If an LLM wants to call into it, it needs to call its local python API.
Is this just unlucky naming or am I missing a critical piece?
MCP is a spec that is attempting to standardize a communication pattern for registering and calling tools from an llm. Part of the spec is a server that exposes specific JSON-RPC end points with a registry of the available tools, resources, and templates, and a way of executing them. That's the server, in this case the server acts as the interface into Blender.
The pipeline for LLM to MCP and to the app looks like,
LLM -> chat app -> MCP client -> MCP server -> specific app (Blender)
The chat app doesn’t know how to talk to Blender. It knows about MCP and links in a client. Blender exposes its functionality via a MCP server. MCP connects the two.
A server-client architecture can run on a single computer. You just need one piece of code to act as the "server" and one to act as the "client". Technically you don't even necessarily have to involve the networking stack, you can just communicate between processes.
tracerbulletx|7 months ago
ww520|7 months ago
camdenreslink|7 months ago