AI usage
How AI agents discover the right brain
When an assistant is connected to several brains, how does it pick the right one? Through brain metadata and intent matching. Here is how discovery and automatic selection work over MCP.
Connect your knowledge over MCP
Connect via MCPA single assistant can be connected to many brains at once - your personal notes, several company brains, a marketplace subscription. So how does it know which one to query for a given question? Not by guessing. It uses each brain’s metadata to match the question to the right source. This article explains how discovery and automatic selection work.
Discovery, step by step
Why metadata is the routing signal
When an assistant lists the brains it can reach, what it actually reads is their metadata - title, description, and intents. That is how it decides which brain fits a question. A specific, well-written description routes cleanly; a vague one leaves the assistant guessing. This is why the description is the highest-leverage thing to get right when you build a brain.
Working with multiple brains
With several brains connected, selection becomes automatic: the assistant picks the brain whose description best matches the question, queries it, and answers. Because each brain is focused, the choice is clean - which is exactly why splitting knowledge into focused brains beats one monolith. It gives the assistant something clear to route between.
Discovery is only as good as your descriptions. If an assistant keeps reaching for the wrong brain, the fix is almost always a sharper description on the brains involved.
Connect your knowledge over MCP
Create or subscribe to a brain on KBrain, then connect its MCP endpoint to Claude, ChatGPT, or any MCP compatible assistant in a few minutes.
Connect via MCPFrequently asked questions
How does an AI assistant know which brain to use?
It reads each connected brain’s metadata - title, description, and intents - and matches the question to the brain that fits, then queries that one. The routing happens on the description before any knowledge is retrieved, so a clear description is what makes selection accurate.
What happens when I have several brains connected?
Selection is automatic. The assistant lists the brains it can reach, picks the one whose description best matches the question, and queries it. Because focused brains have distinct descriptions, the assistant can route cleanly between them.
My assistant keeps picking the wrong brain. What do I fix?
The descriptions. Discovery is driven by brain metadata, so if an assistant reaches for the wrong brain, sharpen the descriptions so each clearly states its domain and the questions it answers. That is almost always the fix.