MuleSoft Agent Fabric: a control plane for agents built anywhere
MuleSoft used Dreamforce 2026 to position Agent Fabric as the answer to agent sprawl and weak AI returns. Here is what it contains and what to ask.
Most large organisations now have AI agents in more than one place: in Salesforce, in Microsoft tools, in home-grown Python projects and in SaaS products that ship their own agents. MuleSoft’s message at Dreamforce 2026 (keynote 16 September) was that this sprawl needs one place to register, govern and connect agents. That place is Agent Fabric.
What Agent Fabric contains
According to MuleSoft, Agent Fabric is built from several parts:
- Agent Registry and Agent Scanners to find and catalogue agents across the organisation, including agents built on other platforms.
- Agent Broker to route work between agents.
- Governance to apply the same policies to all agents.
- AI Gateway and Agent Visualizer for traffic control and an overview of how agents interact.
- Agent Evals, new at Dreamforce, which generates test scenarios, scores results against success criteria and surfaces failures in production.
MuleSoft also stresses cost control: real-time visibility into consumption, hard limits against overruns, routing that matches model cost to task complexity, and tracing tokens back to business value. Agent Fabric supports the Model Context Protocol (MCP), so existing APIs and integrations can be exposed as tools for agents.
Why it matters
MuleSoft frames the launch around a McKinsey figure: 94% of enterprises report that they are not getting strong returns on their AI investments. The argument is that the problem is not the models but the lack of control, context and connection to the systems where work actually happens.
That argument plays to MuleSoft’s strengths. The company has spent years building API management, and Agent Fabric extends the same thinking to a new kind of consumer: policies, gateways and visibility that used to cover APIs now have to cover LLM traffic, MCP servers and autonomous agents too.
How it compares
MuleSoft is not alone. Boomi has Agent Control Tower and an MCP registry in its API management, and Workato has been building agent capabilities into its platform. The difference is in emphasis. MuleSoft builds from API management and governance, with Omni Gateway as the common layer across APIs, agents, LLMs and MCP servers.
For Salesforce customers, MuleSoft has a natural advantage in the link to Agentforce. For organisations with a mixed landscape, the claim that Agent Fabric governs agents “wherever they are built” is the part worth testing.
What to ask before you buy
- Discovery. How do the scanners find agents on other platforms, and what access do they need?
- Policy enforcement. Are policies enforced at runtime through the gateway, or only documented in the registry?
- Evals. Who defines the success criteria, and can test results be part of your release process?
- Cost. Can cost limits be set per business unit or per agent, and how is token use traced to value?
- Licensing. What is included in existing Anypoint subscriptions and what is new?
Agent governance is quickly becoming a separate category in the integration market. Organisations that already use MuleSoft for API management should look at Agent Fabric as an extension of what they have. Others should use the launch as a checklist for what any platform should be able to do.
Sources
This post was written with AI assistance and reviewed by the editor before publishing.