E-BOOK

MuleSoft Agent
Fabric Technical
Overview

Salesforce

Integrations APIs Agents & Tools Automations Runtimes

Payroll MCP

Salesforce Agent Amazon Bedrock

Agentforce

Inventory & Stoc

Amazon Bedro

MCP
Azure AI Foundry Billing MCP Server
Billing MCP Server MCP Employee Onboarding Broker Agent

Azure AI Foundry MuleSoft

Ticketing System Agent HR System Agent Badging

Amazon Bedrock Azure AI Foundry Amazon Bed

IT Broker Agent Talent Pool MCP

MuleSoft MuleSoft
Intranet Manager Agent

Amazon Bedrock

Salesforce Agent
Ping ID Agent
Agentforce
Agent Fabric Technical Overview 2

Contents

Executive Summary 3

Introduction: Shifting from Digital to Agentic Transformation 4

What is MuleSoft Agent Fabric? 5

Technical Foundations of MuleSoft Agent Fabric 6

MuleSoft for Enterprise MCP: From API-Led to Agent-Led 9

Capability Pillars of MuleSoft Agent Fabric 12

Discover 12
Orchestrate 14
Govern 16
Observe 19

Architectural Considerations and Best Practices 21

Customer Journey to Build an Agent Network 23

Conclusion 25
Agent Fabric Technical Overview 3

Executive Summary
The modern enterprise is becoming ‘agentic,’ as a new wave of autonomous AI agents reshape
workflows and promises significant gains in productivity and innovation. Hyperscalers
are creating these agents, SaaS applications are embedding them, and individual teams
are building their own. This explosion of creativity is a powerful force, but it can be difficult
to manage. This places CIOs and IT teams in a stark dichotomy: the board demands they lead
this transformation, yet they face a minefield of security and auditing concerns. Ultimately, opting
out is a false choice. Even without building a single custom agent, organizations will inherit
thousands of independent agents embedded in their SaaS stack, making the governance of this
unmanaged workforce an unavoidable imperative.

We’ve seen this pattern before. Just as the explosion of apps and APIs once threatened
to overwhelm IT with fragmentation, this influx of agents presents a parallel crisis of scale.
MuleSoft previously solved this by helping organizations scale their transformation with a flexible
foundation, and now, MuleSoft extends that two-decade legacy to the new paradigm of multi-
agent systems.

MuleSoft Agent Fabric is the essential solution to help organizations accelerate this innovation.
It provides a unified, open foundation to discover, orchestrate, govern, and observe every
agent and tool across a company’s ecosystem, regardless of where it was built. MuleSoft Agent
Fabric goes beyond a simple control plane: it serves as a full agent management platform that
manages the complete lifecycle of agent-native business processes. It governs diverse elements,
such as agents, MCP Servers, and LLMs, built and run in a heterogeneous environment. By
mitigating key risks while accelerating innovation, MuleSoft ensures every interaction is secure,
compliant, and observable, enabling enterprise architects and IT leaders to regain control over
their agent ecosystems.

Commerce Human Resources Finance

Agent Fabric Technical Overview 4

Introduction: Shifting from

Digital to Agentic Transformation
The business landscape is undergoing a profound transformation, driven by the shift from
a model of connectivity to one of autonomous action. For decades, MuleSoft has been at the
forefront of digital transformation, providing the foundational infrastructure to connect disparate
applications and data sources through reusable APIs and integrations. This era was defined
by the principle of API-led connectivity, which solved the problem of application and API sprawl
by creating a composable enterprise.

Now, a new technological wave is emerging: the agentic enterprise. This agentic transformation
is a fundamental shift in enterprise architecture and operations powered by AI agents that are
capable of autonomous action, real-time adaptation, and complex collaboration. These agents
go far beyond traditional automation or chatbots; they enable an enterprise where human and
agent labor work together to drive key business processes. However, the rapid proliferation of AI
agents across business units, embedded within SaaS products, or built in-house, is giving rise
to a new form of fragmentation known as agent sprawl.

Agent sprawl is a direct parallel to the application and API fragmentation that MuleSoft
was originally built to address. The uncontrolled proliferation of AI agents introduces a web
of operational and architectural challenges that prevent AI from moving beyond experimentation
to scalable, production-grade infrastructure.

The core issues include a lack of a centralized registry, bespoke orchestration between
agents, and a critical absence of robust governance. Without a unified management layer,
each new agent becomes a siloed, fragile, point solution, creating disconnected workflows
and significant compliance risks.

This e-book serves as a strategic guide on how MuleSoft Agent Fabric provides the essential
architectural framework to overcome these challenges and enable a secure, governed, and
scalable agentic enterprise.
Agent Fabric Technical Overview 5

What is MuleSoft Agent Fabric?

To understand MuleSoft Agent Fabric, one must first recognize the fundamental shift in
architectural requirements posed by agentic workloads. The proliferation of disconnected
AI agents poses a significant challenge by hindering orchestration, secure collaboration,
and access to critical data and tools.

MuleSoft Agent Fabric provides a unified foundation for the agentic enterprise, transforming
uncontrolled agents into a governed, highly coordinated, and intelligent network. It enables
organizations to discover, orchestrate, govern, and observe any agent from a single, centralized
platform, ensuring secure, compliant, and business-aligned interactions at scale.

By leveraging components like the Agent Registry for discovery, Agent Broker for orchestration,
Flex Gateway for governance, and Agent Visualizer for real-time observation, Agent Fabric
helps businesses overcome the complexities of ad-hoc solutions and confidently scale their
AI strategies.

Discover Orchestrate Govern Observe

Agent Registry Agent Broker Flex Gateway Agent Visualizer
Agent Fabric Technical Overview 6

Technical Foundations
of MuleSoft Agent Fabric
MuleSoft Agent Fabric is built on the Anypoint Platform, leveraging its proven foundation
for security and connectivity. This approach provides several key advantages, ensuring
universal actionability across any system, data source, or process. The technical architecture
of MuleSoft Agent Fabric is built upon four interconnected capability pillars:

Agent Fabric Reference Architecture

Agent Registry

Agents External MCP Servers Agent AI MCP API

Agent Fabric Broker Connectors Servers Catalog
Components

Governance Policies: MCP (Schema Validation, PII Detector) | A2A (Agent Card, PII Detector) | Rate Limiting | Access Control | Custom Policies

Agent Orchestration
AI Models

Humans Reasoning | Intent Tools/Resources

MCP & Agents
Agent Broker Mule MCP + APIs

AI Agents

Agent Visualizer

Applications

Real-time Map Trace Decision Paths Proactive Insights

Data as Contextual Asset (Storage/Retrieval) MuleSoft Platform Identify & Access Management

Knowledge Graphs Vector DB Hierarchical Trees

Ethics | Compliance | Security | Certs | Monitoring
In House Data 36O SaaS

Discover: The Agent Registry provides

a centralized catalog for all agent assets.

Orchestrate: The Agent Broker acts as

the intelligent routing and execution engine.

Govern: Agent Policies with Flex Gateway

enforces runtime security and compliance with
native support for MCP and A2A protocols.

Observe: The Agent Visualizer offers

real-time topology and AI-specific metrics.
Agent Fabric Technical Overview 7

Built on Anypoint Platform

MuleSoft Agent Fabric is deeply integrated into the Anypoint Platform, leveraging established
enterprise capabilities:

• Anypoint Exchange serves as the repository for the Agent Registry to register and discover agents.
• Anypoint API Manager acts as the central control plane for defining and applying governance
policies enforced by Flex Gateway.
• Anypoint Code Builder (ACB) provides the modern Integrated Development Environment
(IDE) necessary for defining and building Agent Broker logic.
• CloudHub 2.0 and Runtime Fabric (Limited Availability) provide a resilient, scalable runtime
execution and governance environment.
• Flex Gateway provides a high-performance, low-latency control point for APIs and AI agents.

This architectural approach ensures that MuleSoft Agent Fabric provides universal actionability,
empowering agents to act across the entire enterprise IT landscape by connecting proprietary
agent platforms with MuleSoft’s extensive connectivity to systems, data, and processes. The
new agent-native Agent Fabric components leverage updated existing Anypoint capabilities
to provide a unified platform for both digital and agentic transformations.
Agent Fabric Technical Overview 8

Protocol Standardization: The Roles of MCP and A2A

MuleSoft Agent Fabric achieves enterprise-scale interoperability and governance by natively
supporting two critical, emerging open standards: the Model Context Protocol (MCP) and
the Agent-to-Agent (A2A) protocol. It’s important to note that these are initial standardization
attempts, and MuleSoft’s architecture is designed to easily incorporate future standards
as they evolve.

Model Context Protocol

MCP defines how Large Language Models (LLMs) and
agents dynamically discover, invoke, and chain enterprise
tools (APIs, RPA bots, and more) together. By implementing
MCP, traditional APIs are transformed into agent-ready
assets with embedded semantics and metadata, allowing
AI agents to call these tools autonomously. MuleSoft’s
role is to provide the underlying integration backbone
and enterprise governance, ensuring that while MCP
enables autonomy, MuleSoft Agent Fabric ensures that
autonomy is safe, scalable, and governed.

Agent-to-Agent Protocol
The A2A protocol is crucial for enabling secure
communication, collaboration, and coordinated action
among diverse agents across complex enterprise
environments. A2A defines how agents can delegate
tasks and exchange context securely without requiring
access to each other’s internal state, memory, or tools.
To ensure maximum compatibility, A2A supports a
variety of transport protocols, including JSON-RPC
2.0, gRPC, and standard HTTP+JSON, which is essential
for building robust, multi-agent workflows.
Agent Fabric Technical Overview 9

MuleSoft for Enterprise MCP:

From API-Led to Agent-Led
A challenge in the agentic enterprise is the tool selection problem to perform action. LLMs,
trained on natural language, are inherently poor at consuming raw, technical APIs. This mismatch
occurs because traditional APIs are designed for developers, not AI reasoning. They lack the
actionable context and natural language semantics an agent needs to reliably match a user’s
intent (e.g., “Check my order”) to a cryptic endpoint (e.g., GET /v2/status?id=...), leading to poor
tool selection and unreliable outcomes.

MuleSoft Agent Fabric solves this by providing an enterprise-grade framework for the creation,
maintenance, and coverage of MCP servers. MuleSoft customers are uniquely equipped to adopt
MCP, as their integration landscape is already built upon the highly structured and reusable
API-led composable architecture.

This three-layer model provides the perfect foundation for transforming enterprise capabilities
into MCP-compliant assets:

Purpose in Role in
Layer
API-Led Connectivity MCP Enablement

Unlock data from core systems Provide clean, reusable access

System APIs (SAP, Salesforce, databases) points to raw data that can be
by abstracting complexity. exposed as MCP Resources.

Orchestrate data and apply

Directly become the complex,
core business logic across
high-value MCP Tools that
multiple System APIs (e.g.,
Process APIs agents need to execute multi-
creating a 360-degree
step business workflows (e.g.,
customer view or managing
“Process a new sales order”).
order fulfillment).

Serve as an integration point

Tailor data and processes for to be converted into an Agentic
Experience APIs
specific consuming channels. Layer to provide agent-specific
context and actions.
Agent Fabric Technical Overview 10

This existing, layered architecture allows customers to selectively expose the most valuable
business logic as sophisticated MCP tools. A critical step is recognizing that simply pointing
an AI agent at a traditional API is not enough. MuleSoft’s role is to bridge this gap by wrapping
these governed, high-value APIs with the rich, standardized metadata required by MCP. This
transformation enables LLM reasoning and streamlines scale, turning your existing, stable
API foundation into a portfolio of agent-ready capabilities.

Enterprise Concern MuleSoft MCP Feature Business Impact

Agents access live business

data unlocked by APIs through
Accuracy & Trust Real-time integration the MCP layer, preventing
reliance on stale training data
and reducing AI hallucinations.

Enforces the security policies

(OAuth, rate limiting, audit
Governance & Compliance Unified Security Model
trails) on AI agent interactions
as on human-facing APIs.
More about MCP Policies.

The Experience and Process

APIs are wrapped as concise,
high-value Tools with rich,
Capabilities exposed
Address Tool Selection natural language descriptions.
as MCP tools
This drastically improves
the LLM’s ability to select
the correct action.

Agents trigger complex

Process API workflows (e.g.,
Action & Automation Actionable Intelligence automated inventory checks,
logistics scheduling) without
human intervention.

The protocol is open, meaning

the same MCP-enabled asset
LLM, framework works with different LLM
Future-Proofing
and protocol agnostic providers (Anthropic, OpenAI,
etc.), securing the enterprise
from vendor lock-in.
Agent Fabric Technical Overview 11

Creation: Transforming APIs into Agent-Ready Tools

MuleSoft transforms existing portfolios of connected APIs into agent-ready assets
by wrapping APIs, representing business capabilities, with MCP. This process
leverages the core integration backbone to handle complex data transformations
and enables bi-directional communication, instantly providing AI agents with
a library of interactive, enterprise-ready tools.

Maintenance: Centralized Governance and Operations

All MCP servers, running anywhere, are managed through the Anypoint Platform,
with API Manager acting as the central control plane for their entire lifecycle.
This allows you to apply bi-directional policies, which are then enforced by Flex
Gateway. Flex Gateway’s high-performance, native MCP parsing ensures robust
security with minimal latency.

Coverage: A Discoverable and Managed Portfolio

Agent Registry on Anypoint Exchange acts as the single source of truth, providing
comprehensive portfolio coverage for all MCP servers. This ensures all MCP tools
are discoverable, searchable, and properly governed across their entire lifecycle,
empowering both developers and agents to find and reuse the correct, approved
asset for any job.
Agent Fabric Technical Overview 12

Capability Pillars of
MuleSoft Agent Fabric

Discover: Agent Registry

The MuleSoft Agent Registry, built on Anypoint Exchange, is a centralized catalog designed
to surface every AI agent, tool, and asset. This includes assets built in-house and embedded
in SaaS systems.

This capability is fundamental for creating a composable agentic enterprise where agents
and tools are easily discoverable. This empowers developers and other agents to find
and reuse existing assets, which prevents redundant work and accelerates innovation.

Exchange MaxSoft ? AU

Back to AssetBack
List to Asset List

GeoSpatial
GeoSpatial Route Route
Planning Planning Conductor
Conductor Share

Provides advanced route planning, traffic analysis, and custom map generation services. This agent can calculate
Provides advanced route planning,
optimal routes, traffic
estimate travel analysis,
times and real-time
considering custom map
traffic,generation services. This
and create personalized agent
maps withcan calculate
points of interest.
optimal routes, estimate travel times considering real-time traffic, and create personalized maps with points of interest.
Agent Mythical (internal) Last update just now
Agent Mythical (internal) Last update just now

Agent Details Agents

Agents

Instances
Weather Agent
Access List Weather Agent
A helpful agent who can talk to users about the weather.
A helpful agent who can talk to users about the weather.
Agent Mythical (internal) Last update just now
Source File
Agent Mythical (internal) Last update just now

Traffic Agent
A helpful agent who can talk to users about the traffic.
Traffic
Agent
Agent
Mythical (internal) Last update just now
A helpful agent who can talk to users about the traffic.

Tools Agent Mythical (internal) Last update just now

Tools Google Maps MCP

MCP tool for accessing Google Maps.

Agent Mythical (internal) Last update just now

Google Maps MCP
MCP tool for accessing Google Maps.

Agent Mythical (internal) Last update just now

Agent Fabric Technical Overview 13

Key Features and Technical Capabilities

Feature Technical Capabilities

Catalogs agents, MCP Servers, LLMs, and agent

brokers in Anypoint Exchange, creating a universal
Centralized Asset Catalog
repository that makes valuable enterprise assets
easily discoverable and reusable.

Defines and registers assets within your enterprise’s

Declarative Registration
unified agent network.

Allows teams to catalog and govern offline

Offline Asset Support or legacy systems by supporting the upload
of an A2A-compliant agent card.

Manages distinct endpoints and credentials for

development, test, and production environments
Environment Management
in API Manager, simplifying promotions
between stages.

Leverages native Anypoint Exchange features like

Enhanced Search
tagging and categorization to help developers and
and Categorization
agents quickly find the most relevant tool for a job.
Agent Fabric Technical Overview 14

Orchestrate: Agent Broker

The MuleSoft Agent Broker is a context-aware routing service that dynamically matches tasks
to the most suitable agent or tool. It also serves as an organizational component for your agentic
enterprise, allowing you to group agents and MCP servers into business focused domains.

For example, a Broker can manage an HR Domain, routing incoming HR tasks across X agent(s)
and X MCP server(s), to achieve the optimal outcome.

With a dedicated routing service that enables agents to work together, organizations can build
tailored agents, focused on tackling specific tasks, rather than “super agents” that attempt
to tackle it all.

This increases the accuracy of responses while still allowing agents to address more complex
problems that span multiple systems and services.

human-resources

MULESOFT DEV AGENT [Link] [Link]

our HR System. [Link] [Link]

54 tools:
Dev Agent has a question: 55 — mcp:
Dev Agent has a question: Agent Network
Human Resources 56 server: talent-pool-mcp
We found the HR System agent. Would Agent Network
you like to use this agent? Human Resources 57 allowList:
We found the HR System agent. Would 58 — talentPoolMCPServer.match_email_to_address
you like to use this agent? 59 — mcp:
Yes Broker
Employee Onboarding Broker 60 server: slack-mcp
Broker 61 allowList:
Yes Task Completed Employee Onboarding Broker 62 — ConductorSlackMcpServer.send_status_update
I’ve added the HR system agent Agents MCP Servers
63 links:
your agent network! 64 — salesforce-agent:
Task Completed Agents MCP Servers
65 authentication:
Dev Agent has a question:
I’ve added the HR system agent Agent MCP Server
66 oAuth2ClientCredentials:
your agent network!
What LLM would you like to use? Salesforce Talent Pool 67 policies:
Einstein, OpenAI, or Llama? 68 inbound:
Agent MCP Server 69 — tracing
OpenAI Salesforce Talent Pool 70 — guardrails
71 outbound:
Task Completed Agent MCP Server 72 — tracing
HR System Slack 73 — guardrails
Okay. I’ve updated your broker 74 — hr-agent:
to use OpenAI LLM.
75 authentication:
Agent MCP Server 76 oAuth2ClientCredentials:
Add Ticketing System and IT agents. Also HR System Slack 66 policies:
the Talent Pool and Slack MCP Servers.
Agent 78 inbound:
Ticketing System 79 — tracing
80 — guardrails
I added the following assets. No changes 81 outbound:
were made to your LLM, MCP servers, 82 — tracing
or existing Salesforce/HR System agents. Agent
83 — guardrails
ticketing-agent Ticketing
Agent System 84 — ticketing-agent:
badging-agent IT Broker 85 authentication:
it-broker-agent
86 oAuth2ClientCredentials:
talent-pool-mcp
slack-mcp 87 policies:
88 inbound:
89 — tracing
Type your prompt... Agent
90 — guardrails
IT Broker
91 outbound:
Agent Fabric Technical Overview 15

Key Features and Technical Capabilities

Feature Technical Capabilities

Implements the Agent Loop pattern, which

Dynamic Orchestration Pattern
dynamically delegates tasks to the best-fit agent.

Utilizes an LLM to interpret natural language

LLM-Powered Reasoning goals, delivering intelligent, context-aware
outcomes without requiring complex, rigid logic.

Allows developers to select specific LLM

Configurable LLM
models in the agent network project, providing
Model Selection
flexibility to balance cost, performance,
and advanced capabilities.

Using MuleSoft Vibes, developers can define

agent broker’s logic with natural language, boosting
AI-Powered Development
productivity and making it easier to build complex
agent orchestrations.

Automatically applies transaction tracing

upon deployment, providing instant, out-
Automated Observability Hooks
of-the-box visibility into agent performance
and simplifying debugging.

Deploys as a containerized application to Mule

Managed Cloud Deployment Runtime, which reduces operational overhead
and ensures high availability and scalability.
Agent Fabric Technical Overview 16

Govern: Flex Gateway for A2A/MCP

MuleSoft Agent Governance provides the essential security and governance layer for the
agentic enterprise by extending the proven, enterprise-grade Flex Gateway to natively support
MCP and A2A interactions. As autonomous agents handle sensitive data and execute critical
business processes, a lack of centralized governance poses an unacceptable risk.

MuleSoft Agent Governance allows organizations to enforce consistent security policies,

manage access, and monitor every agent interaction from a single, unified platform. This
transforms ungoverned agent sprawl into a secure, compliant, and trustworthy network.

Anypoint API Manager serves as the central control plane for defining and applying policies
to every agent in the network. Each agent, LLM, MCP Server, or Agent Broker is represented
as a managed asset within API Manager, allowing administrators to deploy a Flex Gateway
to enforce policies and create a secure perimeter for all traffic. Flex Gateway allows you to apply
MCP and A2A policies that govern traffic to agents, agent brokers, MCP servers, and APIs.

These policies can be applied to both incoming (ingress) and outgoing (egress) traffic for agents
and agent brokers, creating robust controls around their operations.

APIs Automations Network Marketplace Mythical

API Manager
Agent Administration / GeoSpatial Route Planning Conductor / Add policy
SANDBOX
Search by name

Agent Administration
Browse Show all policies
Policies
A2A
Logs

Settings Agent card policy Agent Content Logging Agent Rate Limiting
Rewrite Agent card to point Log all content through Apply Rate limiting to individual
to backend API. MCP servers. A2A tools & resources.
Learn more Learn more Learn more

Agent Spike Control Agent Prompt Decoration Agent PII Policy

Apply Spike Control to individual Modify the incoming prompt with Protect sensitive personal data
MCP tools & resources. context as input to A2A server. to identify an individual
Learn more Learn more Learn more

Cancel Previous Next

Agent Fabric Technical Overview 17

Key Out-of-the-Box Policies

Policy Application to Governance

Extends existing API Management

capabilities to MCP servers, enabling policies
MCP Support
like authentication, rate limiting, and caching
to be applied to AI-callable tools.

A critical guardrail that prevents accidental

A2A PII Detector leakage of sensitive customer or employee
data by inspecting payloads.

Enforces the A2A specification on all requests,

A2A Schema Validation ensuring that the agent only receives and sends
well-formed, compliant messages.

Automatically enriches prompts that the

A2A Prompt Decorator Broker sends to other agents or LLMs with
critical context (e.g., security tokens, user IDs).

A configuration that rewrites the Agent Card URL

to match the public URL of the gateway instance,
A2A Agent Card
ensuring other agents receive the correct, governed
endpoint address.

Policies like JWT validation, client ID enforcement,

Rate Limiting and spike control are applied to requests coming
and Access Control into the Broker to protect it from unauthorized
access and DoS attacks.
Agent Fabric Technical Overview 18

Features and Technical Capabilities

Feature Technical Capability

Manages all policies from API Manager, providing

Centralized Policy
a single control plane that ensures consistent security
Configuration
standards are applied to every agent and API.

Parses and validates A2A and MCP traffic, allowing

Native A2A/MCP for the application of fine-grained security policies
Policy Enforcement that understand the unique structure and risks
of agent communication.

Provides out-of-the-box policies like the A2A PII

Specialized A2A/MCP Detector, which accelerates the implementation
Policy Library of critical security measures and protects
sensitive data.

Extends governance beyond out-of-the-box policies

by enabling teams to build their own custom policies
Custom Policy
using the Policy Development Kit (PDK). This allows
Development
for the enforcement of unique, business-specific
logic directly at the gateway.

Built on a high-performance engine, Flex

High-Performance / Gateway secures agent traffic with minimal
Low-Latency Core performance impact.

Captures detailed analytics on agent traffic

Usage Analytics in API Manager, providing valuable insights into
and Reporting usage patterns and helping to accurately manage
LLM-related costs.
Agent Fabric Technical Overview 19

Observe: Agent Visualizer

The MuleSoft Agent Visualizer provides real-time insights into agent behavior, addressing
the ‘black box’ problem often associated with enterprise AI adoption.

As autonomous agents orchestrate complex tasks, their decision-making processes can

become opaque, hindering troubleshooting, performance optimization, and organizational
trust. The Agent Visualizer tackles this by transforming complex, multi-agent interactions
into an intuitive, real-time map of the entire network.

This provides end-to-end visibility, allowing architects and operators to trace decision
paths, identify performance bottlenecks, and monitor key AI metrics, ultimately building
confidence in AI systems.

The Agent Visualizer renders a real-time graph of all brokers, agents, and MCP servers, providing
a complete, easy-to-understand map of the entire agent ecosystem and its complex dependencies.

Unlike traditional process orchestration that relies on defined paths between tasks, agentic
orchestration is probabilistic. Agent Visualizer can notionally display the agents and tools that
could be called, allowing users to visually identify emergent or unexpected interactions as they
occur in real-time.

Salesforce

Integrations
Agents APIs Agents & Tools Automations Runtimes Mythical

Shipping Assistant Agent

Payroll MCP

Salesforce Agent Weather Agent Amazon Bedrock

Anypoint

Agentforce A helpful agent who can talk to users about the weather.

Agent Mythical (internal) Last update just now Agent

Inventory & Stock
Amazon Bedrock

Shipping
MCP
Traffic Agent Oracle
Azure AI Foundry Billing MCP Server MCP Employee Onboarding Broker Agent
Customs & Compliance Agent Agent
A helpful agent who can talk to users about the traffic.
Azure AI Foundry MuleSoft Amazon Bedrock

Agent Mythical (internal) Last update just now

Procurement & Suppliers Agent

Tools Amazon Bedrock

Ticketing System Agent HR System Agent Badging Agent

Amazon Bedrock Azure AI Foundry Amazon Bedrock

Google Maps MCP

MCP tool for accessing Google Maps.
Tax Services MCP
Agent Mythical (internal) Last update just now Amazon Bedrock
IT Broker Agent Talent Pool MCP

MuleSoft MuleSoft Salesforce Agent

Intranet Manager Agent
Agentforce
Amazon Bedrock

Finance Coordinator Agent

Salesforce Agent
Ping ID Agent Amazon Bedrock
Agentforce
Amazon Bedrock
Agent Fabric Technical Overview 20

Features and Technical Capabilities

Feature Technical Capability

Renders a real-time graph of all brokers, agents,

Holistic System
LLMs, and MCP servers, providing a complete, easy-
Topology View
to-understand map of the entire agent ecosystem.

Differentiates between design-time and

Design vs. Runtime runtime relationships, allowing teams to instantly
Visualization see if agent behavior in production matches
its intended design.

Displays key operational metrics like task

Granular
failures and MCP success rates, enabling teams
Transaction Metrics
to proactively identify performance bottlenecks.

Provides direct links from any node to its logs,

Integrated Log
dramatically reducing the mean time to resolution
and Trace Navigation
(MTTR) for errors and system failures.

Displays a shield icon on governed agents,

Gateway Security
providing an at-a-glance confirmation that
Status Indication
security policies are active and enforced.
Agent Fabric Technical Overview 21

Architectural Considerations
and Best Practices
The diagram illustrates a simple multi-agent system architecture, for an order management
process. An Agent Broker acts as the central hub, receiving requests from the internet via
a CloudHub Ingress.

These requests pass through a Flex Ingress Gateway governed by policies, which then directs
them to an Agent Broker performing orchestration for the Order Management component.
This orchestrator is responsible for coordinating various tasks by communicating with
different specialized agents.

The orchestrator’s output is routed through a Flex Egress Gateway, also subject to governance
policies, to specific agents like the Status & Tracking Agent or the Order Agent. These agents
interact with various MCP servers, such as the ERP Core MCP Server (on MuleSoft) and Shipping
MCP Server (running outside MuleSoft), to perform their designated functions.

For instance, the Order Agent might trigger a refund_process tool on the Billing MCP Server.
An Agent Registry manages and lists all available agents, MCP servers, and LLMs (deployed
anywhere), while an Agent Visualizer provides a visual map of the agent network.

Overall, the system uses a distributed, agent-based approach to manage complex order workflows.

Agent Registry

ERP Core Finance / Billing Customer

Order Agent
MCP Server MCP Server MCP Server
BYO-LLM
Agent Fabric Status and Shipping
Heroku AI LLM Refund Agent
Tracking Agent MCP Server
Components

CH 2.0 (Private Space)

Agent Broker
ERP Core MCP Server
CH 2.0 (Private Space)
order_status
Status and
check_eligibility
Tracking Agent
status_update
Agent Policies
Agent Policies

Agent Policies

Agent Policies

CloudHub Flex Ingress Agent Orchestration Flex Engress

Internet Ingress Gateways for Order Management Gateways
Finance / Billing
MCP Server
refund_process
Order Agent
Customer MCP Server
contact_details

Agent
Agent Visualizer Policies
Policies

Policies
Agent

Agent

Loyalty Mgmt Agent Refund Agent

Anypoint Monitoring Shipping MCP Server
carrier_tracking
Agent Fabric Technical Overview 22

MuleSoft Agent Fabric is not a standalone product but a new layer built directly on the
existing MuleSoft Anypoint Platform, which provides a proven foundation for API management
and integration. This architectural approach provides several critical advantages, including:

• Universal Actionability: MuleSoft Agent Fabric leverages MuleSoft’s universal connectivity to

empower agents to act across any system, data source, or process.
• Deployment Flexibility: The platform offers a range of deployment options to meet the diverse
needs of modern enterprises, including strict compliance requirements and data residency
laws. MuleSoft Agent Fabric can be deployed in CloudHub or Runtime Fabric (Limited
Availability), offering the flexibility to run the platform near agentic endpoints for minimal
latency. This is particularly important for regulated industries like finance and healthcare.

The platform’s architecture ensures that organizations can extend their existing security
and governance policies to a new class of assets, leveraging a trusted foundation to move
into the agentic era with confidence.

In an agentic enterprise, the Agent Fabric functions as a critical layer to manage the design
time and runtime interactions between agents, tools, and LLM providers. It plays a central role
in managing reliability, observability, and governance for every request flowing into production.

Because the Agent Fabric manages production traffic, it must be deployed with the following
enterprise-grade quality-of-service requirements in mind:

• High availability to enable global deployments with regional fault isolation.

• Fault tolerance so that the nodes do not become a single point of failure
and continue serving traffic gracefully.
• Low latency to ensure minimal overhead and a good customer experience.
• High throughput and scalability to handle thousands of concurrent
requests with efficient resource usage.
• Separation of control plane and runtime plane so that configuration
changes are decoupled from production traffic routing.
• Security and privacy to extend existing security and governance policies
to these new assets.

MuleSoft Agent Fabric addresses these requirements through its robust integration
with the Anypoint Platform, leveraging established capabilities for security, scalability, and
high availability. Its distributed architecture and flexible deployment options ensure fault
tolerance and low latency, while the separation of control and runtime planes guarantees
efficient management of production traffic.
Agent Fabric Technical Overview 23

Customer Journey to
Build an Agent Network
Adopting MuleSoft Agent Fabric is not a monolithic, one-size-fits-all process; it is designed
to meet organizations where they are. Some might start by simply cataloging and governing
existing agents, while others might begin by exposing their existing APIs as discoverable
tools for other agentic systems.

The table below outlines a logical progression to building an agent network from initial asset
cataloging to observing a fully governed, multi-agent orchestration in production.

Step Action

Catalog your existing APIs, services, MCP

servers, and agents in the Agent Registry.
Register Assets
This makes them discoverable, reusable
assets for any future orchestration.

Build your Agent Broker project in Anypoint Code

Builder by defining a business goal and leveraging
Configure the Agent Broker the agents you’ve already registered. MuleSoft
Vibes can even help you create your project with
natural language.

Before deployment, apply security policies

to the Agent Broker and individual agents using
Govern the Network
API Manager/Flex Gateway. This ensures security
and compliance across the entire interaction.

Deploy the Agent Broker and use the Agent

Visualize Interactions Visualizer to monitor the live agent network,
tracking performance and troubleshooting any
issues that arise.
Agent Fabric Technical Overview 24

This end-to-end journey is just one of several ways to leverage Agent Fabric. Organizations
can adopt its capabilities incrementally, layering on sophistication as their needs evolve. Below
are seven distinct adoption patterns, from foundational enablement to advanced orchestration.

Wrap APIs with MCP: The simplest entry point is to take existing REST APIs
and wrap them with the MCP. This instantly makes them LLM-callable, turning
any API into a discoverable “tool” that an AI agent can call using natural language.

A2A-Enable Existing Agents: Standardize agents built on any platform, even those
created before the A2A protocol existed. By wrapping non-compliant agents, you bring
them into the ecosystem, allowing them to communicate using a standard protocol.

Curate Agents & Tools: Use the Agent Registry to catalog and curate an official,
approved set of agents and MCP tools. This design-time governance is crucial for
establishing a single source of truth and ensuring developers and agents use only
vetted, compliant assets.

Govern Agents & Tools: Move from design-time to runtime governance by placing
a Flex Gateway in front of your registered agents and tools. Apply out-of-the-box A2A
or MCP policies to secure interactions, or develop custom policies to handle unique
business logic and compliance needs.

Visualize Agent Interactions: Use the Agent Visualizer to see how your governed
agents and tools interact. This provides an initial view of your agent network,
allowing you to understand dependencies, trace interactions, and analyze
the impact of individual components.

Perform Agent Orchestration: Create an Agent Broker in Anypoint Code Builder to

orchestrate multi-step processes. This can start with an agent broker orchestrating
a set of predictable tools and can later evolve to a more advanced agent broker that
leverages both deterministic and non-deterministic external agents.

Visualize the Entire Fabric: Evolve from visualizing individual interactions to

observing the entire, holistic agent fabric. This comprehensive view in the Agent
Visualizer shows how agent brokers, agents, and tools all connect and collaborate,
allowing you to manage and optimize the complete, interconnected agentic enterprise.
Agent Fabric Technical Overview 25

Conclusion
MuleSoft Agent Fabric represents the essential infrastructure layer required for enterprises
to transition autonomous AI from isolated experiments into scalable, governed, and integrated
production components.

By unifying discovery, dynamic orchestration, stringent runtime governance, and real-time

observation under the proven Anypoint Platform, MuleSoft Agent Fabric strategically mitigates
the severe risks posed by agent sprawl. The architecture ensures that every agent interaction
is secure, compliant, observable, and aligned with core business objectives, empowering
organizations to become resilient, high-performing, and truly interconnected agentic enterprises.

Discover how MuleSoft Agent Fabric Prepare your organization for agentic
helps organizations orchestrate and transformation to deliver highly
govern every AI agent. personalized experiences at scale.

Explore now Read the white paper

The information provided in this report is strictly for the convenience of our customers and is for general informational purposes only.
Publication by Salesforce, Inc. does not constitute an endorsement. [Link] does not warrant the accuracy or completeness of any information,
text, graphics, links, or other items contained within this guide. [Link] does not guarantee you will achieve any specific results if you follow any
advice in the report. It may be advisable for you to consult with a professional such as a lawyer, accountant, architect, business advisor, or professional
engineer to get specific advice that applies to your specific situation.

© Copyright 2025, Salesforce, Inc. All rights reserved.