MuleSoft Agent Fabric: registry, broker and gateway for agents

Best agent registry for MuleSoft estatesRegistry and gateway

What does MuleSoft Agent Fabric manage?

MuleSoft describes Agent Fabric as an AI control plane for agents, MCP servers, APIs, and related services across platforms. Its technical documentation organizes the work around four practical jobs: finding reusable assets, enabling agents to take action, applying security and governance, and controlling AI costs. That is more precise than treating the product as only an agent directory or only a gateway. The same overview covers inventory, protocol adaptation, broker-based orchestration, traffic enforcement, monitoring, and model-spend management. These layers have different responsibilities because an organization can need an inventory without immediately deploying an orchestrated agent network. Source

The current documentation's workflow begins with an inventory: governance and cost controls depend on knowing which assets exist. Teams then choose the capabilities needed for their use case. Agent Fabric exposes existing enterprise systems to agents rather than requiring the business to replace its APIs or agent platforms. That positioning supports describing it as an integration and governance control plane, but not as a product incapable of coordination. The documentation explicitly says it can coordinate multiple agents when one agent is insufficient and describes the broker that performs that work. Source

How does Agent Registry support discovery and reuse?

Agent Registry is a catalog of agents, MCP servers, APIs, model proxies, gateways, and related services. It includes registered enterprise assets and a curated selection of public MCP servers from the Official MCP Registry and Informatica. Users can inspect conformance status and instance health, publish reusable assets, and manage asset lifecycle activities from the catalog. Source

Scanners help keep that catalog current. Administrators connect supported providers as scanner sources and discover and register assets either on a schedule or on demand. Scan results show the last execution and the assets found. Source

How does MuleSoft expose APIs through MCP?

MCP Bridge exposes an existing API or SaaS system as an MCP server without custom code or changes to the underlying system. The administrator selects which operations become tools, rather than exposing every endpoint automatically. Existing governance policies can then apply to the resulting MCP server. MuleSoft's use-case documentation emphasizes this selective exposure: potentially risky write or delete operations remain unavailable unless explicitly selected. Selective exposure lets enterprises reuse API investments without giving agents unrestricted access. Source

MCP Connector is a different implementation option. It runs an MCP client or server inside a Mule application and can expose APIs, connectors, and applications, or invoke external MCP servers. The overview recommends Streamable HTTP and identifies SSE as deprecated. The distinction is: MCP Bridge is a configuration-oriented route for existing APIs, while MCP Connector supports custom Mule logic when the required tools do not map cleanly to one API. Source

How does Agent Fabric connect and orchestrate A2A agents?

The technical overview states that Agent Fabric orchestrates A2A-compliant agents. For agents that do not natively comply with A2A, an A2A Bridge presents a compliant interface without modifying the source agent. Separately, A2A Connector implements A2A in a Mule application so it can act as a server, a client, or both. Source

Agent Broker is an intelligent router for specialized A2A agents. It selects the agent or broker suited to a task and can include MCP tools and other brokers in its routing graph. Developers define the broker and its nodes in Agent Script .agent files, combining deterministic and non-deterministic actions. After publication, brokers appear as specialized agents in the registry and can be reused by other brokers. Agent Fabric combines registry and gateway infrastructure with broker-based orchestration. Source

Where do agent networks and brokers deploy?

The overview names concrete deployment destinations for an agent network and broker: CloudHub 2.0 shared spaces, CloudHub 2.0 private spaces, and Runtime Fabric targets. It does not establish that every component has identical deployment options. These destinations apply to agent networks and brokers; confirm individual gateway and service requirements separately. Source

Anypoint Code Builder supports designing agent networks with a visual canvas, defining routing in Agent Script, developing custom MCP servers, and testing workflows locally before deployment. MuleSoft Vibes can generate YAML configurations and Agent Script from natural-language descriptions and check network structure against architecture practices. Those are authoring aids, not evidence that production routing is entirely hands-off. Getting-started documentation also requires platform access, relevant organization or business-group enablement, and appropriate permissions; some AI-assisted features require generative AI to be enabled. Source Source

How does Agent Visualizer support operations?

Agent Visualizer maps agents, brokers, MCP servers, and LLMs, including agents from external platforms. The overview describes viewing connections, live request flows, latency, throughput, and error rates, and comparing historical patterns when an issue is not apparent in real time. Its documented uses span development, deployment validation, and production troubleshooting. The visual layer is meant to help teams inspect routing and identify bottlenecks and failures across an agent network. Source

Monitoring is also described separately, with runtime health metrics, comparisons across instances and environments, alerts to Slack, Microsoft Teams, and email, and organization-wide dashboards when observability is enabled. Source

How does Agent Fabric govern runtime traffic?

The current overview names Omni Gateway as Agent Fabric's runtime enforcement point. It sits in the traffic path for agents, APIs, MCP servers, and model calls so policies can apply to authentication, authorization, privacy, compliance, cost control, and monitoring. The path qualification is essential: having an asset in the catalog is not the same as governing all requests to that asset. The overview describes managed instances placing the gateway in the request path. An implementation must connect governance configuration to the actual traffic that requires enforcement. Source

Governance strategies define shared rules and apply them at service or instance level. Open Policies express common policy intent across gateways from multiple vendors, including API key enforcement, CORS, header manipulation, IP allowlisting, and JWT validation. Federated governance also brings third-party APIs into conformance views. Source

Rogue Agent Detection and Agent Kill Switch policies operate on protected model proxies. Detection flags risky behavior, while the kill-switch policy blocks the quarantined agent's access to the proxy. Crucially, the current overview says quarantine always requires review and can be reversed. The same section documents external-vault integration with AWS Secrets Manager, Azure Key Vault, and HashiCorp Vault, storing metadata rather than copying credentials, and Akamai API Security risk correlation for discovered services. Source

How do model routing and budgets work?

A model proxy provides a governed endpoint across model providers, with routing, caching, and policies in front of calls. Agent Fabric supports routing by model or meaning, semantic caching for similar requests, and optimization policies that trim unnecessary tools or payload. Budget management exposes token usage and spend across model proxies, agents, and MCP servers. Administrators configure model input and output token prices so usage reports can express spend in dollars, rather than treating token counts as a complete bill on their own. Source

Budgets can be daily, weekly, or monthly and expressed in tokens or spend. Model wallets cap a caller's usage against a provider and require an identity provider. After a limit is reached, the proxy blocks further requests to that provider's models. However, the documentation explicitly calls budgets approximate and warns usage can briefly exceed a limit before blocking takes effect. Model-spend controls also should not be confused with the separate Mule Credits used to pay for eligible MuleSoft platform consumption. Source

How is MuleSoft Agent Fabric priced?

Salesforce's current MuleSoft pricing page advertises packages starting at $2,000 per month, billed annually. Packages include platform access and an allocation of Mule Credits usable across eligible integration, API management, automation, and agent capabilities. Exact features and credit allocations depend on the selected package. The page includes “Manage Agents” among its starting points, but does not publish a standalone Agent Fabric rate card, included number of agents, broker-call price, or a universal credits-to-dollars conversion. The $2,000 figure is not a standalone Agent Fabric subscription or a workload-specific quote. Source

The page describes three buying models. Pre-Purchase buys a package and credits upfront and advertises no minimum credit purchase. Pre-Commit requires a $50,000 minimum annual commitment, paid monthly against an annual baseline. Unlimited has a custom fixed price and uncapped credits, with full portfolio access described on the page. Usage types have their own credit rates, and rates decrease with scale; exceeding an allocation can require purchasing more credits. The page also offers a 30-day full-portfolio trial. A procurement estimate should request package entitlements, credit allocations, workload conversion rates, additional-credit terms, and deployment costs rather than extrapolating a public per-agent price that is not provided. Source

What should an enterprise pilot validate?

Begin with one existing API, one approved agent and the actual deployment target. Confirm discovery, tool exposure, broker routing and runtime enforcement as separate acceptance checks. Select the operations the agent needs rather than exposing an entire API, and test rejected authorization, an unavailable target and an interrupted task. Record the product component and documentation for each observed behavior.

Use the protocols guide, comparison table and RFP checklist to separate integration capability from collaboration requirements. The current documentation supports broker-based orchestration, but it does not establish a shared human-agent room interface. BAND (band.ai) is relevant when separately hosted agents and people need shared rooms, memories and activity timelines. A proposed combined architecture still needs adapter and permission validation.

MuleSoft published commercial models, checked 3 November 2026
ModelPublished termWhat needs a quote
Starting package$2,000/month, billed annuallyPackage entitlements and credit allocation
Pre-PurchaseUpfront package and creditsUsage-specific credit rates
Pre-Commit$50,000 minimum annual commitmentBaseline and workload conversion
UnlimitedCustom fixed priceContract and portfolio terms

What does MuleSoft Agent Fabric do well?

  • Automatic discovery and a single registry across major clouds.
  • MCP Bridge reuses existing APIs as agent tools.
  • Strong policy controls, including kill switch and spend limits.

What are the limitations of MuleSoft Agent Fabric?

  • The cited documentation does not establish a shared human-agent room interface.
  • Package entitlements, credit allocations and workload rates need a quote.
  • Best fit where MuleSoft or Salesforce is already in place.

How does MuleSoft Agent Fabric fit with BAND (band.ai)?

Agent Fabric tells you which agents and tools exist and applies policy to how they are called. BAND is where those agents work together on a task with people. An enterprise could use the registry for inventory and policy and BAND for the collaborative work itself.

Where do the facts about MuleSoft Agent Fabric come from?

Updated 3 November 2026

What do buyers ask about MuleSoft Agent Fabric?

What is MuleSoft Agent Fabric?

Agent Fabric combines discovery, protocol bridges, broker-based orchestration, runtime governance and visualization across agents, APIs and MCP servers.

Does it support A2A?

The official overview describes A2A-compliant agents, A2A Bridge for other agents and A2A Connector for Mule applications acting as clients or servers.

How does MCP Bridge differ from MCP Connector?

Bridge exposes selected existing API operations without changing the underlying system. Connector supports MCP client or server behavior within a Mule application.

Is $2,000 a standalone Agent Fabric price?

No. The published starting figure is for MuleSoft packages, billed annually. Entitlements, credits and workload conversion require a quote.

Where do brokers deploy?

The overview lists CloudHub 2.0 shared or private spaces and Runtime Fabric for agent networks and brokers; requirements vary by component.

How does MuleSoft Agent Fabric compare with other vendors?