MuleSoft Agent Fabric
by MuleSoft (Salesforce)
A control plane to register, route, govern and cost-cap AI agents and MCP tools from any vendor
MuleSoft Agent Fabric is Salesforce's control plane for finding, connecting, governing and paying for AI agents across an enterprise, including agents built on Microsoft, Google, AWS or Databricks. It is for integration and platform teams dealing with agent sprawl who need one registry, policy layer and kill switch for every agent.
MuleSoft Agent Fabric is an agent management layer built into Salesforce's MuleSoft Anypoint Platform. It launched in September 2025 with four parts: Agent Registry, a catalog for agents, tools and MCP and A2A servers; Agent Broker, a routing service that groups agents into business domains and sends tasks to the best fit using the customer's chosen LLM; Agent Governance, which applies security and policy controls through Flex Gateway for MCP and A2A traffic; and Agent Visualizer, a live map of the agent network. Agents stay where they run and the registry points to them, so customers do not need Agentforce to use it. Since launch MuleSoft has added scanners that detect agents and MCP servers on Amazon, Google, Microsoft, Claude, Databricks and Kong, an MCP Bridge that exposes existing APIs as MCP tools, a unified LLM and MCP gateway, a FinOps dashboard with Model Wallet budgets, and a gateway kill switch that became generally available in September 2026 and can stop a request, session, agent or tenant. In October 2026 Salesforce said Agent Evals and API Evals enter beta in November 2026, with general availability and an evaluation-triggered kill switch planned for January 2027. Pricing runs through Mule Credits, with packages from $2,000 a month billed annually and a $50,000 annual minimum on the pre-commit option. Analysts at Moor Insights and HyperFRAME Research describe it as an extension of MuleSoft's API integration business into agent coordination, competing with Microsoft Agent 365, ServiceNow AI Control Tower and Boomi.
An enterprise integration or platform team, often already on MuleSoft Anypoint, that has agents running on several vendors and no central inventory, policy or spending control for them
One registry, gateway and kill switch that covers agents and MCP servers across Salesforce, Microsoft, Google, AWS and Databricks without moving them
At a Glance
- Category
- Governance & Security
- Pricing
- Subscription, Usage-based, Contact for pricing
- Target Market
- CIOs, CTOs, Enterprise Architects, Integration Teams
- Deployment
- Cloud-first, Hybrid, Multi-cloud
Key Features
- ✓Agent Registry and scanners
Catalogs agents, tools and MCP and A2A servers, and scanners detect agents on Amazon, Google, Microsoft, Claude, Databricks and Kong so nothing stays invisible.
- ✓Agent Broker
Routes multi-step work across agents, tools and people by business domain, using the customer's chosen LLM and a mix of fixed rules and model judgment.
- ✓Unified LLM and MCP gateway
Applies security, compliance and rate policies to every agent, model and tool call, including Open Policies that span Azure, Apigee and Kong gateways.
- ✓Agent kill switch
Stops agent activity at the request, session, agent or tenant level; the gateway version has been generally available since September 2026.
- ✓MCP Bridge
Turns existing APIs into MCP tools so agents can use systems a company has already integrated through MuleSoft without new connectors.
- ✓AI cost controls
A FinOps dashboard, Model Wallet budgets and model routing by task and cost let teams cap spend per agent or team.
Capabilities
Use Cases
- •Agent sprawl audit
A CIO's team runs the scanners across cloud accounts and SaaS platforms to build a single inventory of every agent and MCP server in use.
- •Cross-vendor order workflow
Agent Broker routes a customer request from an Agentforce service agent to a Databricks data agent and a legacy ERP API exposed through MCP Bridge.
- •Production incident containment
An operations team uses the kill switch to halt one tenant's misbehaving agent session without taking down the rest of the agent network.
- •Token budget enforcement
Finance and platform teams set Model Wallet budgets per business unit and route routine tasks to cheaper models to keep LLM spend predictable.
Ideal For
Best For
- ✓Building an inventory of agents and MCP servers already running across multiple clouds and SaaS platforms
- ✓Exposing existing REST APIs to agents as MCP tools without rewriting them
- ✓Enforcing one set of security and rate policies on LLM, MCP and A2A traffic through a shared gateway
- ✓Setting token budgets per team or agent and stopping a misbehaving agent in production
- ✓Organisations already paying for MuleSoft Anypoint that want agent governance on the same credit pool
Not Ideal For
- ✗Small teams or startups, since packages start at $2,000 a month billed annually and reviewers of the wider Anypoint platform repeatedly cite cost and pricing complexity
- ✗Buyers who need agent evaluation in production today, because Agent Evals is only reaching beta in November 2026 with general availability planned for January 2027
- ✗Teams that want an open-source, self-hosted agent registry with no dependence on a closed commercial platform
Deployment
Market Analysis
Pros
- ✓Vendor-neutral registry and discovery across the major agent platforms
- ✓Reuses MuleSoft connectors and APIs that many large enterprises already run
- ✓Production controls such as a scoped kill switch and token budgets are already generally available
- ✓Analysts at Moor Insights describe it as a unified control layer where rivals offer piecemeal tools
Cons
- ✗Reviewers of the underlying Anypoint platform on PeerSpot and rfp.wiki repeatedly cite high cost and complex pricing as the main drawback
- ✗Setup and the DataWeave language carry a learning curve that weighs on smaller teams, per aggregated reviews
- ✗Several headline capabilities, including Agent Evals, API Evals and the evaluation-triggered kill switch, are beta or not due until January 2027
- ✗Customer evidence for Agent Fabric specifically is thin: the product page names one customer, and VentureBeat's October 2026 piece named none
Pricing
Pre-Purchase package
From $2,000/mo (billed annually)
- ✓Buy a package and a block of Mule Credits upfront
- ✓No minimum credit purchase
- ✓Lowest per-credit rate
Pre-Commit
Contact for pricing
- ✓$50,000 minimum annual commitment
- ✓Pay monthly as credits are drawn down
- ✓One shared pool across teams
Unlimited
Contact for pricing
- ✓Custom fixed annual price
- ✓Uncapped credits
- ✓Full portfolio access
Agent Fabric is billed through Mule Credits, a shared currency across MuleSoft integration, API management, automation and agent features, with per-usage credit rates that fall with volume. Packages start at $2,000 a month billed annually; Agent Fabric itself has no separate public price and its agent-management package requires talking to sales.
Security & Compliance
Sources
This page was written from 7 sources, 5 on domains other than mulesoft.com.
- 1.mulesoft.com — agent fabricvendor
- 2.mulesoft.com — pricingvendor
- 3.cio.com — mulesoft launches agent fabric to tackle agent sprawl and un
- 4.venturebeat.com — salesforce details cost controls and kill switch
- 5.peerspot.com — mulesoft anypoint platform vs tray ai
- 6.rfp.wiki — mulesoft anypoint platform
- 7.hn.algolia.com — search
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Goodfire
Interpretability platform for debugging AI models and guardrailing agents with low-cost activation probes
Asymptote Beacon
Open-source endpoint telemetry for AI coding agents, with a paid platform for agent policy and governance
Classie Supervise
Real-time supervision for sanctioned and shadow AI agents, with OPA policies enforced inline across browser, endpoint and cloud
Mave
Agentic SecOps platform that investigates every alert across your existing security tools without ingesting the data