What’s an "agent", and why is everyone saying it?
Agents and MCP · Builder · 5 min · agents, llm · after Next-token prediction
An agent is a language model run in a loop: think, call a tool, read the result, think again, until the task is done. MCP is a shared socket standard so any tool plugs into any model.
1 Guess first
2 Look
The loop
1 / 4"Book me a table for four on Friday near the office and email everyone."
ThinkI need the office address.Use a tool🔌 file readerObserve12 Nguyen Hue, District 1.🔌 web search🔌 calendar🔌 calculator🔌 file reader🔌 emailturn 1 of 4 · 🔌 = a socket the same shape for every tool (MCP)
An agent is a language model in a loop. First it thinks: what do I need next?
The model never acts; it chooses a socket, a tool acts, and the result comes back as text for the next turn.
Each turn the model writes two things: a thought and a tool call — a bit of structured text naming a tool and its arguments. Software runs the tool and pastes the result back into the window. MCP (Model Context Protocol) is the agreed shape for that socket, so a tool built once works with any model that speaks it.
4 You are the model
You are the agent. For each turn, pick the tool that fits the thought, then Go. Four turns to book the table.
"Book me a table for four on Friday near the office and email everyone."
Think: I need the office address.
You are the agent: pick the tool for this turn, then Go.5 Check
6 Say it back
7 Mission · 5 min
Pick a free coding agent from the directory, give it a two-minute task in a scratch folder ("list the files and write a README"), and watch the tool calls go by. Then plug aifree’s MCP server into any client and ask it for free LLM APIs.
Three quick ones
Drawn at random from this lesson's pool; different next time.
Done?
Sources
- https://modelcontextprotocol.io/introduction
- https://www.anthropic.com/research/building-effective-agents
Published Oct 20, 2026 · updated Oct 20, 2026