Getting Started with FaburAI

FaburAI governs the Model Context Protocol traffic between your AI systems and your enterprise tools. It does not sit in the request path: policy is evaluated by an enforcer running in your own cluster.

Four steps take you from nothing to a governed, audited AI call.

1. Register an MCP server

Start in AI Connect → MCP Server Mesh. Register the endpoint your AI workloads already call. FaburAI needs the endpoint URL and whatever credential it takes to authenticate.

You register MCP servers that exist. FaburAI does not ship vendor adapters — if a vendor publishes an MCP server, you point FaburAI at it with your own credentials and FaburAI governs the traffic.

2. Let discovery run

On connect, FaburAI inspects the server and records what it exposes: tools, resources, prompt templates, and reachable data assets. You do not enumerate these by hand.

The results land in AI Catalog. This inventory is what policies target — you cannot govern what you have not discovered.

3. Write a policy

Go to AI Control → Policy Catalog. A policy says which subjects may perform which actions against which targets, under which conditions.

New MCP servers default to observe mode: every call is recorded, nothing is blocked. That is deliberate. Watch real traffic first, then move to simulate to see what your policies *would* do, and only then to enforce.

4. Watch a decision

Make a call through the gateway and open AI Control → Decisions Log. Every governed invocation produces a decision row: who called what, when, and the verdict — with the policies that matched and the one that won.

Where to go next

• AI Graph shows the whole mesh as a graph — actors through servers, tools,
and data — with each hop coloured by policy outcome.
• AI Cockpit shows runtime behaviour over time: activity, policy hit rates,
and token consumption.
• Deploy & Operate covers running FaburAI yourself, including air-gapped
installations.