AI Connect
Discover what is already running.
Most AI estates were not designed — they accumulated. AI Connect starts from that reality: point it at what you have, and it builds the picture for you, then keeps it current.
You cannot govern what nobody has written down.
Agents get built by different teams on different timelines. Servers get stood up for one project and quietly reused by three more. Asking people to fill in a registry produces a document that is accurate on the day it is written.
AI Connect does not ask you to declare your estate. It reads it — from the servers themselves, from the data systems behind them, and from the traffic arriving at them.
One server at a time, and it explains itself.
Point FaburAI at a server — reachable over the network, or running as a local process — and it asks the server what it offers. The protocol already carries that answer, so nobody has to write it down twice.
No manifest to maintain
The server is the source of truth about itself. A description someone typed into a wiki starts drifting the moment it is saved.
Credentials stay protected
Connection credentials for the systems behind a server are encrypted at rest rather than sitting in plain configuration.
Nothing moves
Servers stay where they are, owned by whoever owns them today. Registration is a governance act, not a migration.
Six things, none of them hand-keyed.
MCP servers
Registered over the network or as a local process. FaburAI reads the server’s own description of itself rather than relying on a separate manifest someone has to maintain.
Tools, resources and prompts
Everything the server exposes, including what each tool does and whether it reads, writes or deletes — which is what policy later binds to.
Tables and columns
The data systems behind the server are introspected down to the column, so governance can reach the level data owners actually care about.
Sensitivity signals
Column names are scanned as they are discovered and tagged PII, PHI, GDPR, SOX or HIPAA, so a policy can refer to a category instead of a list.
AI actors
The agents and applications calling in, identified from the provenance on their requests and from the handshake they open the connection with.
Roles
Synced from the enterprise identity provider you already run, so policy binds to roles your organization has already defined.
When a server changes, that is an event.
Each time FaburAI examines a server it records a fingerprint of what that server exposes — its tools, its resources, its prompts, its capabilities. Comparing fingerprints is what turns change into something you can see.
A tool added quietly to a production server is a detectable event, not something you find out about during an incident.
New capability appears
A server starts exposing an operation it did not expose last week. Whether that is intended is a question worth asking on the day it happens.
Something disappears
A tool that policy referred to is gone. Rules written against it are now governing nothing, quietly.
The estate can be reconstructed
Because snapshots are kept rather than overwritten, the state of a server during a past period can be shown rather than described.
Agents nobody registered show up anyway.
Every client that connects to a governed server identifies itself when it opens the connection. FaburAI records that identity, and any caller it has not seen before becomes an entry in the catalog.
Observed
Seen calling, and not reviewed. This is where a previously unknown agent arrives.
Curated
A person has looked at it, confirmed what it is, and attached the context the automated signal could not supply — ownership, purpose, team.
Approved
Accepted as a known part of the estate. Policy can reference an actor’s state, so a rule can treat reviewed and unreviewed callers differently.
The practical effect: the registry maintains itself, and the gap between what is documented and what is running stays small. An actor that has been calling for weeks without anyone reviewing it is visible as exactly that.
Discovery runs on a schedule, not once.
Per-server intervals
How often a server is examined is set on that server. A fast-moving internal service can be checked frequently while a stable vendor server is left alone.
Reachability
Servers are probed on an interval, so one that has stopped responding shows up as a health signal rather than as silence.
A tool added next month, or a column added next quarter, is picked up without anyone re-running an import.
See what discovery finds in your estate.
If you are not sure what is currently calling what, that is the normal starting position and the one AI Connect is built for.