MuleSoft launches Agent Fabric to tackle agent sprawl and unify enterprise AI workflows

Salesforce-owned integration platform provider MuleSoft has launched Agent Fabric, a new suite of capabilities designed to help enterprises tackle the growing challenge of agent sprawl.

Agent sprawl, often a result of enterprises and their technology teams adopting multiple agentic products lured by low-cost offers from vendors and concerns of lock-in, according to HyperFRAME Research analyst Stephanie Walter, can lead to agents and their workflows being fragmented, redundant, or siloed across teams and platforms.

This sprawl, Walter warned, is highly likely to threaten operational efficiency and complicate governance, making it harder for enterprises to scale AI responsibly.

Agent Fabric with tools, such as Agent Registry, Agent Broker, Agent Visualizer, and Agent Governance, will try to tackle agent sprawl by providing a “single place” to register, orchestrate, govern, and observe every deployed agent, regardless of the platform it was built, Andrew Comstock, SVP and GM of MuleSoft, said.

The Agent Fabric will be integrated into MuleSoft’s Anypoint platform.

How the tools inside Agent Fabric work

In order to start using Agent Fabric, enterprises will need to bring all their agents and tools, including MCP and A2A servers, into one place using the Agent Registry tool.

Bringing all the tools in one place makes them discoverable by other agents and developers, the company said, adding that this should make each asset easy to find, reuse, and compose into workflows.

Once enterprises have brought all their agent-related assets together, they can use the Agent Broker tool, an intelligent routing service powered by Salesforce’s Atlas Reasoning Engine, to connect agents across domains, dynamically matching tasks with the best-fit agent.

Under the hood, the Broker tool connects disparate agents using MCP and A2A, and enterprises also get the choice of using their preferred LLM as the base for the service, the company said.

Post determination of the multi-agent workflow, enterprises can use Agent Governance and Agent Visualizer for governance and observability, respectively.

While Agent Governance can be used to apply guardrails to ensure security, compliance, and policy controls for all agent interactions, the Agent Visualizer tool can be used to view a dynamic map of the agent ecosystem deployed, including how they are connected and performing, the company said.

Agent Registry, Agent Broker, and Agent Visualizer are expected to be made generally available in October. Agent Governance is available now.

Replicating APIs’ success with Agent Fabric

Analysts feel that MuleSoft is looking to extend its API integration legacy with Agent Fabric as enterprises continue to adopt AI agents for automation.

“Just as APIs required gateways and management, agents now need coordination, governance, and observability,” Robert Kramer, principal analyst at Moor Insights and Strategy, said.

MuleSoft, according to Kramer, aimed to help solve challenges around APIs integration with its Anypoint Platform and is now trying to cut through piecemeal agent governance and observability solutions that other vendors offer with Agent Fabric by providing a unified control layer across multiple agents.

However, analysts also pointed out that other vendors, such as Boomi, Microsoft, Google, and AWS, are also working towards offering capabilities that tackle agent sprawl.

“Most vendors are recognizing that the central challenge for AI adoption at scale is not just building individual agents but managing the complexity of a sprawling network of agents that need to work together securely and reliably. They are all developing fabric-like or foundry-like platforms to provide this critical orchestration and governance layer,” HyperFRAME’s Walter said.

Seconding Walter Moor Insights and Strategy’s Kramer pointed out that each vendor is playing to their strengths: Microsoft offers Fabric and MCP for agent orchestration, Google is embedding agents into Vertex AI and BigQuery, IBM is expanding orchestration with watsonx, and Databricks and Snowflake are developing agentic features around their data ecosystems.