Get more AI into production without losing control of what it can reach or do. Agent Fabric applies one set of policies across every agent, MCP server and model in your enterprise, without any of them being rebuilt, migrated or instrumented.
Core controls are authored in one place and pushed to every gateway you already run, including AWS, Azure, Apigee and Kong. No translation step in between, which is where environments usually drift apart.
Identity stays attached through every hop rather than being verified once and assumed for the rest. Permissions are checked continuously and scoped to what the task actually needs.
The kill switch severs an agent at the network, credential and identity layer at once, so it loses access to everything it could reach in a single action.
Agents are being built on Bedrock, Vertex, Copilot, Agentforce and by your own engineers, each with its own tools and controls. Agent Fabric sits across all of them, so platform and security teams govern from one place instead of one platform at a time.
Existing production APIs become agent-usable tools through the governance layer you already run, with no rewrites, no new runtimes and no trade-offs.
Teams pick which approved API operations to expose and map them to MCP tools in minutes, without touching the services behind them. Pre-built SaaS actions cover 100+ common apps including Jira, Asana, HubSpot and Zendesk.
Authentication, rate limiting and observability already applied to your APIs extend automatically to agent traffic, with no per-use-case configuration.
A single endpoint can safely span different systems without embedding orchestration logic into the agents themselves, so governance holds as adoption scales.
Security, compliance and observability apply at the protocol layer, bidirectionally, across every agent-to-agent interaction. Agents that don't natively speak A2A are bridged in a few clicks.
Identity propagates through every interaction, so nothing acts outside its permitted scope no matter how many agents a workflow passes through.
Every agent-to-agent and agent-to-application interaction is logged and auditable, giving operations teams the context to debug, optimize and demonstrate compliance.
Kill switch alerts land in email or Slack as soon as an agent is flagged, with the triggering signal and its evidence attached, so you can tell a real problem from a false positive.
Most enterprises govern AI three separate times, once for APIs, once for models and once for agents. Incidents happen in the gaps between them. One governance layer closes those gaps.
Open Policy Catalog centralizes native, out-of-the-box policies to AWS API Gateway, Azure API Management, Apigee and Kong. A REST API on Apigee, an MCP server on Azure and an agent on Omni Gateway can be overseen from the same place and enforced in the same way.
Guardrails from AWS, Azure and F5 Calypso extend to agent traffic rather than being rebuilt for it. Akamai covers shadow APIs and MCP servers, and secrets stay in HashiCorp, AWS or Azure, wherever you already keep them.
Rate limiting, PII detection and masking, prompt protection and data access controls apply to OpenAI, Anthropic, Gemini and Bedrock alike, through one governed endpoint your developers authenticate to once.
Explore this deep dive on MuleSoft Agent Fabric to learn how to orchestrate and govern a secure network of enterprise AI agents.
Brief description that explains what this resource covers and why people should check it out. Focus on the value they'll get from it.
See how IT leaders are approaching integration, automation, and AI adoption in 2026 — and where the gaps between ambition and execution are widest.
Try MuleSoft Anypoint Platform free for 30 days. No credit card, no installations.
Tell us a bit more so the right person can reach out faster.
Get the latest news about integration, automation, API management, and AI.
Policy is applied at the communication layer, so a third-party agent is governed without being rebuilt, migrated or instrumented. That is what makes Agent Fabric vendor agnostic: the same governance applies whether an agent was built in Salesforce, on a hyperscaler, or inside software you bought.
Rate limiting, authentication, PII detection and masking, and data access controls, enforced on any agent regardless of the platform it was built on. Policies are written once and pushed to the gateways you already run, including AWS API Gateway, Azure API Management, Apigee and Kong.
Agent Fabric governs the three at one layer instead of three. Model calls, MCP tool access and agent-to-agent communication all run through Omni Gateway, so the same policy set and the same audit trail apply to all of them.
The agent kill switch enforces constraints at the network, credential and identity layer at the same time, cutting an agent off from everything it could reach in one action. You see the flag history and the signal that triggered it before you decide, and alerts arrive in email or Slack the moment it happens.
MCP support exposes existing APIs as agent-ready tools at the gateway layer, with no backend modifications. Security policies, access controls and audit logging are inherited automatically, and you choose which endpoints agents can reach.
Agents built on AWS Bedrock, Google Vertex AI, Microsoft Copilot, Azure AI Foundry, Agentforce and your own engineering teams, alongside the models behind them including OpenAI, Anthropic, Gemini and Bedrock.
