mulesoft-agentfabric

star 0

MuleSoft Agent Fabric / Omni Gateway / Enhanced MuleSoft Experience specialist — the AI-and-agents layer that sits ON TOP of classic Mule integrations. Covers the Agent Registry/Exchange catalog, Agent Broker (deterministic orchestration), Omni Gateway (MCP/A2A/LLM governance, formerly Flex Gateway), LLM Proxy, Agent Visualizer, Trusted Agent Identity, building agent networks in Anypoint Code Builder, and the MuleSoft → Salesforce Agentforce bridge. Use when exposing Mule APIs as agent-ready MCP tools, governing agent/LLM traffic, or building agentic workflows. NOT for classic integration (use /mulesoft-flow, -connector, -dataweave) or platform admin of standard Mule apps (use /mulesoft-platform).

Mrsavage92 By Mrsavage92 schedule Updated 6/10/2026

name: mulesoft-agentfabric description: "MuleSoft Agent Fabric / Omni Gateway / Enhanced MuleSoft Experience specialist — the AI-and-agents layer that sits ON TOP of classic Mule integrations. Covers the Agent Registry/Exchange catalog, Agent Broker (deterministic orchestration), Omni Gateway (MCP/A2A/LLM governance, formerly Flex Gateway), LLM Proxy, Agent Visualizer, Trusted Agent Identity, building agent networks in Anypoint Code Builder, and the MuleSoft → Salesforce Agentforce bridge. Use when exposing Mule APIs as agent-ready MCP tools, governing agent/LLM traffic, or building agentic workflows. NOT for classic integration (use /mulesoft-flow, -connector, -dataweave) or platform admin of standard Mule apps (use /mulesoft-platform)."

Skill: MuleSoft Agent Fabric (AI & Agents Layer)

Purpose

The AI-layer specialist for the /mulesoft suite. Agent Fabric is MuleSoft's agentic control plane — discover, govern, orchestrate, and observe AI agents, MCP servers, and LLM traffic from one place. It is the front-end you saw at eu1.omni.mulesoft.com (the Enhanced MuleSoft Experience).

This skill is for the AI/agent work. The classic build skills (-flow, -connector, -dataweave, -platform) remain the foundation Agent Fabric consumes.

Read This First — Foundation Before Agents

Agent Fabric does not replace integration skills — it consumes them. The marketing line "your APIs become agent-ready tools in minutes" only holds if the APIs underneath are clean, governed, and deployed. BDR's classic System/Process API work (salesforce-system-api, broadband lookups) is the substrate the agent layer plugs into. Do not pause classic integration to chase agents; agents need those APIs to exist first.

Sequence: build clean APIs → register them in Exchange → expose as governed MCP tools → let agents (Agentforce or external) consume them.

TWO HARD GATES before any agentic build

  1. Entitlement gate. BDR's tier is "Customer – Integration Advanced" (see reference_bdr_anypoint_tier memory). Agent Fabric / Omni Gateway / LLM Proxy are newer paid add-on capabilities that may NOT be in that SKU. Same discipline that stopped the custom-monitoring build: do not plan an Agent Fabric build until BDR's entitlement is confirmed with the MuleSoft account team (Anil / account exec). VERIFY — do not assume included, do not assume excluded.
  2. Use-case gate. "We'll use AI agents" is a direction, not a spec. Refuse to generate agent networks or broker config until there is a concrete task an agent should perform (e.g. "Agentforce agent runs broadband availability lookups in natural language"). Generic agent scaffolding is banned.

When to Use

  • Exposing existing Mule APIs as MCP servers / agent-ready tools
  • Registering agents, MCP servers, or LLMs in Exchange / Agent Registry
  • Governing MCP / A2A / LLM traffic via Omni Gateway (policies, rate limits, identity)
  • Routing LLM calls through LLM Proxy (OpenAI, Gemini, Bedrock/Claude, NVIDIA)
  • Building an agent network in Anypoint Code Builder
  • Connecting Mule APIs/actions to Salesforce Agentforce
  • Observing agent interactions via Agent Visualizer

When NOT to Use

  • Building/deploying a standard Mule app → /mulesoft-flow, /mulesoft-platform
  • Connector auth, DataWeave, classic flow error handling → respective subskills
  • BDR project status/blockers → /mulesoft-bdr
  • Anything where entitlement or use-case gate is unmet → STOP, surface the gate

The Agent Fabric Stack

Enhanced MuleSoft Experience (eu1.omni.mulesoft.com)
├── Anypoint Exchange / Agent Registry  ← catalog of agents, MCP servers, LLMs
├── Agent Broker                        ← intelligent A2A router; deterministic orchestration GA Jun 2026
├── Omni Gateway (was Flex Gateway)     ← governs MCP / A2A / LLM traffic, policies, identity
├── LLM Proxy                           ← one endpoint for OpenAI/Gemini/Bedrock(Claude)/NVIDIA
├── Agent Visualizer                    ← visual map + traces of agent interactions
└── Trusted Agent Identity              ← identity propagation; mobile auth for high-risk actions

Classic Anypoint Platform (eu1.anypoint.mulesoft.com)  ← UNCHANGED, still the foundation
└── Code Builder/Studio · Runtime Manager · CloudHub 2.0 · API Manager · Secrets Manager

Key fact (verified): the Enhanced Experience is complementary, not a replacement (docs). Classic build/deploy (Code Builder, CI/CD, CloudHub 2.0, Exchange, API Manager) is unchanged. Agent networks themselves deploy to CloudHub 2.0 — the same runtime BDR already uses.

Components (detail in references/components.md)

Component What it does What you build/configure
Exchange / Agent Registry Central catalog of AI assets Register/publish APIs, MCP servers, agents; discover public MCP servers
Omni Gateway Flex Gateway + MCP/A2A/LLM governance Federated policies, OpenAPI→MCP conversion, token mgmt, correlation-ID tracing
LLM Proxy Unified multi-provider LLM endpoint Model routing, centralized token/cost mgmt, provider failover
Agent Broker A2A router that plans + delegates Intent/policy/identity-based routing; deterministic orchestration (GA Jun 2026)
Agent Visualizer Observability for agent networks Visual map; rides on Anypoint Monitoring metrics/logs/traces
Trusted Agent Identity Identity for agent actions Identity propagation A2A & agent-to-app; mobile auth for high-risk actions

Modes

Mode When Reference
Learn the landscape New to Agent Fabric (current state) references/components.md
Expose an API as MCP tool Make a Mule API agent-ready references/omni-gateway-and-governance.md
Govern agent/LLM traffic Apply policies, identity, rate limits references/omni-gateway-and-governance.md
Build an agent network Author + deploy in Code Builder references/agent-networks.md
Connect to Agentforce Mule APIs as Agentforce actions references/agentforce-bridge.md
Strategic gates Before any build references/foundation-and-entitlement.md

Honesty Markers — what is verified vs needs doc confirmation

This skill was built from official docs + announcements (June 2026). Some build-level specifics were not exposed in public docs and are explicitly flagged in the references rather than fabricated:

  • Verified: component purposes, Enhanced-vs-classic complementarity, CloudHub 2.0 deploy target, LLM Proxy providers, GA dates (AI Gateway/MCP Bridge/Trusted Agent Identity GA; deterministic orchestration beta Apr 2026 → GA Jun 2026).
  • NOT verified (do not invent — confirm against docs or a live tenant): exact agent-network YAML field names/schema, Omni Gateway policy property names, BDR's Agent Fabric entitlement/SKU, Agentforce integration step-by-step.

When a reference says "confirm in docs," fetch https://docs.mulesoft.com/agent-fabric/ or the linked guide — do not guess field names.

Proactive Triggers

  • User asks to build an agent network / broker without a concrete use case → STOP, surface use-case gate.
  • User asks to enable Omni Gateway / LLM Proxy without confirming entitlement → STOP, surface entitlement gate (pairs with feedback_check_subscription_before_build).
  • User wants to expose an API as an MCP tool but the API isn't deployed/governed yet → route to classic skills first (foundation-before-agents).
  • User conflates Omni Gateway with classic API Manager → clarify: Omni Gateway = Flex Gateway extended for MCP/A2A/LLM; classic REST API policies still live in API Manager (/mulesoft-platform).
  • User assumes Enhanced Experience replaced Anypoint Platform → correct: complementary, classic workflow unchanged.

Anti-Patterns (do NOT do these)

  • Fabricating agent-network YAML schemas or Omni Gateway policy names — public docs were thin; confirm against docs/tenant.
  • Planning an agentic build before entitlement is confirmed.
  • Generic "AI agent" scaffolding with no concrete task.
  • Treating Agent Fabric as a replacement for integration skills — it's a layer on top.
  • Routing LLM calls direct to providers from a Mule app when LLM Proxy exists (loses token mgmt, governance, failover).
  • Exposing an ungoverned API as an MCP tool (no auth/rate limit = open agent attack surface).

Related Skills

  • /mulesoft — orchestrator; routes here for AI-layer work
  • /mulesoft-platform — classic deploy, API Manager policies, monitoring (the foundation)
  • /mulesoft-flow, /mulesoft-connector, /mulesoft-dataweave — build the APIs agents consume
  • /mulesoft-bdr — BDR project state; AI is Phase 6 of the enterprise roadmap
  • /sync-knowledge-base — run after editing this skill
Install via CLI
npx skills add https://github.com/Mrsavage92/claude-config --skill mulesoft-agentfabric
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator