The universal socket — MCP.
Last lesson, a soloist called in specialists. But hand-wiring every specialist to every band doesn't scale. MCP is the universal socket: build a tool's plug once, and any model can use it.
In Lesson 9 a model reached past itself to call a tool — a metronome, a session log, an arranger. Useful, until you look at the wiring. Each tool needed its own custom cable, built around one model's particular way of talking to that tool. Add a second model, a third tool, a fourth, and every new pairing demands another hand-made cable. The tangle grows fast.
MCP — the Model Context Protocol — is the fix, and it works exactly like a universal socket. Just as a quarter-inch plug lets any guitar reach any amp, MCP is one agreed-on shape. A tool builds a single standard plug once, and every MCP-speaking app can use it — whether that app is a chat client, a coding tool, or an autonomous agent.
That's the whole win, and it comes down to counting. Without a standard, joining a handful of models to a handful of tools takes a cable for every pairing — models times tools. With the universal socket, each side wires up just once — models plus tools. The interactive makes that gap jump out.
Bands on the left, specialists on the right. Flip between hand-made cables and one universal socket, then add a band or a specialist and watch how the wiring count grows in each world.
Build the plug once.
Without a standard, every model–tool pairing is its own integration. With MCP, a tool is an MCP server built once, and any MCP client (inside an AI app) can connect to it.
# before: a bespoke adapter for every pairing — models × tools connect(bandA, metronome, adapter="bandA↔metronome") connect(bandA, session_log, adapter="bandA↔session_log") connect(bandB, metronome, adapter="bandB↔metronome") # ...and on, and on # with MCP: each side speaks one standard, so it just plugs in metronome = MCPServer(tools=["bars", "click"]) # built once bandA = MCPClient() # speaks the standard bandA.connect(metronome) # any client ↔ any server
One socket, everything plugs in.
MCP is a shared standard for how tools and AI apps talk. A tool exposes itself as a server one time; any app that speaks the protocol can use it — no custom cable per pairing. That turns a models × tools tangle into a tidy models + tools, which is what lets a whole ecosystem of ready-made connectors exist for you to just plug in.
Lesson 9 gave the model the ability to call a tool. This is the plumbing that makes calling any tool practical — the universal socket behind the whole thing.