Can we skip the gateway and call the model from the agent?

You can. You will not be able to prove who spent the tokens or which policy applied. We will not sign that as production.

Is Flex Gateway the same as Agent Fabric?

No. Flex Gateway is the edge. Agent Fabric is the control plane that inventories agents, MCP, and policies. You usually want both.

MuleSoft Flex Gateway

Every tool call gets a front door.

The first week is in the org, reading what already exists. Flex Gateway is where identity, policy, and traces live. Agents do not get a shared key. The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.

Get the written assessment

Certified Partner since 2010 · MVP Hall of Fame · 200+ agents in production · UAE and US desks

Trusted identity

Jobs

What the gateway is for.

Identity

The user authenticates once. Every hop exchanges.

01
User authenticates
The identity provider issues the first token. The channel is Slack, a portal, or Salesforce.
02
Gateway exchanges
OAuth on-behalf-of credential injection. The next audience is narrower than the last.
03
Broker and agents
The broker routes. Downstream agents and MCP servers see a user, not a shared service account.
04
Trace
Which identity, which policy, which tool. That is observability, not a log dump.

Trusted agent identity: the user signs in. Each Flex Gateway hop exchanges for a narrower token scoped to the next audience. The agent does not run as a god-mode connected app unless you choose that, and we will say when that is a demo.

Mermaid source

flowchart LR
  U[User] --> C[Channel]
  C --> G1[Flex Gateway]
  G1 --> B[Broker]
  B --> G2[Flex Gateway]
  G2 --> A[Agent]
  G2 --> M[MCP]
  U -. authenticates .-> I[Identity provider]
  I -. OBO token .-> G1
  I -. OBO token .-> G2

MCP

Catalog, entitle, then serve.

The gateway is the front door for tools. Listing a server is not permission.

AI Gateway

Model calls belong on the same door as tools.

Routing, budgets, and guardrails in one place. Not three SDK configurations and a hope.

01
Route by job
High-stakes judgement to a frontier model, classification and transforms to a fast one. Written down, not decided per engineer.
02
Meter per caller
Token accounting per business group, so a runaway loop is a throttle instead of an invoice.
03
Guardrails at the edge
Denied topics, PII detection, and response checks on the hop rather than inside each prompt.
04
Keep the choice reversible
A model swap should be a gateway change. If it is a rewrite, the routing was in the wrong place.

Model routing through the gateway. One entry point in front of frontier reasoning, fast high-volume models, open models inside your own perimeter, and dedicated forecasting models.

Mermaid source

flowchart TD
  A["Agents and applications"] --> G["AI Gateway: identity, policy, budget, audit"]
  G --> F["Frontier reasoning"]
  G --> H["High-volume extraction"]
  G --> L["Open models in your perimeter"]
  G --> T["Time-series forecasting"]

Questions

What we actually say.

Can we skip the gateway and call the model from the agent?: You can. You will not be able to prove who spent the tokens or which policy applied. We will not sign that as production.
Is Flex Gateway the same as Agent Fabric?: No. Flex Gateway is the edge. Agent Fabric is the control plane that inventories agents, MCP, and policies. You usually want both.

The brief · one email

Show us what the agent can call today.

We will mark the hops with no identity, and the ones that should not exist.