Core Concepts
The problem: Agent sprawl
Specialized AI agents are distributed across teams, functions, and ecosystems. Without a scalable way to find and track these agents, organisations end up with a black box of ungoverned agents — trapped in silos, leading to one-and-done solutions that can’t be reused or governed.
What are Agent Scanners?
Agent Scanners, part of MuleSoft Agent Fabric, automatically find and catalog AI agents from external providers directly into the MuleSoft Agent Registry — eliminating manual data entry.
Supported providers include:
| Provider | Platform |
|---|---|
| Amazon | Amazon Bedrock Runtime |
| Salesforce | Agentforce |
| Vertex AI | |
| Microsoft | Copilot Studio |
In this workshop you’ll connect to Amazon Bedrock and watch the scanner automatically discover and register agents from AWS.
Why centralise all agents in one registry?
Unified governance and security
When agents are scattered across platforms, it’s nearly impossible to enforce a consistent security standard. With agents cataloged in Agent Registry, you apply a single set of enterprise policies — like PII detection or access control — through Omni Gateway, regardless of where the agent lives.
Enable Agent-to-Agent (A2A) collaboration
The most powerful AI workflows involve agents collaborating with other agents to solve complex business tasks. For example, a customer support agent might collaborate with a Shipping Agent to track a package. MuleSoft Agent Broker intelligently routes requests to the right agent in the registry — turning point-to-point connections into a governed agent network.
Agile tech stack
Modern AI moves fast. You might switch platforms to leverage a better model, reduce latency, or take advantage of more competitive pricing. Agent Scanners let you do this without losing visibility or governance — the registry stays current regardless of which platform each agent runs on.
Real-time observability
A centralised registry coupled with Agent Visualizer turns agent sprawl into a visual map. You can see which agents are used most, which are reused, and where the bottlenecks are — all from a single control plane.
How Agent Scanners work
External Platform Agent Scanner Agent Registry
(Amazon Bedrock) │ │
│ │ 1. Authenticate │
│◀────────────────────────│ │
│ │ │
│ 2. List & fetch agents │ │
│────────────────────────▶│ │
│ │ 3. Create/update │
│ │ asset records │
│ │────────────────────────▶│
│ │ │
│ On schedule: repeat │
Each scanner runs on a configurable schedule and syncs changes automatically — new agents are added, updated agents are refreshed, and removed agents are flagged. No manual import required.
What gets imported
When the scanner discovers an agent, it imports:
- Agent name and description
- A2A agent card — the metadata that tells other agents how to call this one
- Source platform — which provider the agent lives on
- Alias / version — the specific deployed version discovered
Continue to Add Amazon Bedrock Scanner →