StoryboardAI Engineering

Model Context Protocol

A standard plug between assistants and tools, so each tool is connected once and works everywhere.

What is the Model Context Protocol?

Every tool an assistant uses needs wiring: a description for the model, code to make the real call, handling for errors. Written by hand, that wiring is different for every system, and has to be repeated in every product that wants the tool.

The Model Context Protocol, MCP, replaces that with one common language. A system is wrapped in a small MCP server. Any assistant that speaks MCP can connect, ask what tools are on offer, and call them in the same standard way.

The saving is in the arithmetic. Three products and three systems need nine custom connections, or six standard adapters. And an adapter someone else wrote works with your assistant unchanged.

MCP: one standard way to offer tools to any assistant.Step 1: A system publishes a server. Step 2: The harness asks what tools it has. Step 3: Tools join the model's list. Step 4: Requests pass through the harness.1A system publishesa server2The harness askswhat tools it has3Tools join themodel's list4Requests passthrough theharness
MCP: one standard way to offer tools to any assistant.

Why does the Model Context Protocol matter?

Before a shared standard, every assistant needed custom code for every system it used. Ten assistants and ten systems meant a hundred integrations. The Model Context Protocol, MCP, is a common plug: a system describes its tools once, and any assistant that speaks the protocol can use them. It is why assistants can now be connected to new systems in minutes.

How does the Model Context Protocol work?

  1. The makers of a system publish a small program called an MCP server.
  2. Your harness connects to it and asks what tools it offers.
  3. The server answers with each tool's name, description and inputs, in a standard shape.
  4. Those tools are added to the list the model sees, alongside your own.
  5. When the model asks for one, the harness passes the request to the server and returns the result.

When should you use the Model Context Protocol?

Use it when

  • You want an assistant to reach a system someone else built
  • The same tools should work across several assistants
  • A provider already publishes a server for their product

Think twice when

  • You need one simple tool inside your own code. A plain function is less to maintain
  • You cannot vouch for the server. Its descriptions and results enter your model's window

Where does the Model Context Protocol sit in a real system?

MCP sits between your harness and outside systems. The harness is the client. Each connected system is a server. The model itself never speaks the protocol. It sees the resulting tools exactly as it sees any others.

What goes wrong with the Model Context Protocol?

  • Trusting a server's tool descriptions as instructions. They are text written by a stranger.
  • Connecting many servers and filling every prompt with hundreds of tool descriptions.
  • Giving a connected tool the same freedom as your own without separate permission rules.

What does the Model Context Protocol look like in practice?

A coding agent needs the team's issue tracker. The tracker's maker publishes an MCP server. The harness connects, asks for its tools and receives four: get an issue, search issues, add a comment, close an issue. The two that write are set to ask a person first.

Watch it happen

These stories show model context protocol at work, one step at a time.