Back to the canvas
Architecture
LLM, agent and MCP
How a question travels: an agent holds the model and reaches your systems through MCP.
What it shows
Three terms that get used as if they were one. The LLM is the model that reads and writes text. The agent is the model plus the loop that lets it take steps. MCP is the adapter through which the agent reaches your systems.
The diagram places each one, and shows where your own systems sit: outside the agent, behind the adapter.
Key decisions
- An agent is not a bigger model: it is a model with tools and a loop.
- Access to your systems goes through one adapter, which is where permissions belong.
- Your systems stay outside the agent: it can use them, it does not contain them.