Blazing fast AI-native agent gateway

LLMs, MCP servers, APIs, and agents, routed, governed, and observed through a single endpoint deployed inside your own cloud.

50+ providers1,200+ modelsLLM, MCP, A2A and API supportPostman-hosted or hybrid deployment

Fabric Gateway is an AI-native gateway built for agents, LLMs, MCP servers, and the systems they interact with.

Why Fabric Gateway

The enterprise AI adoption problem

What breaks when AI adoption expands across teams—and how Fabric restores control.

Fragmented AI access

Teams integrate models, MCP servers and internal APIs independently, creating inconsistent access and duplicated infrastructure.

With Fabric: Connect every AI service through one governed gateway.

Unreliable providers

Outages, latency spikes and rate limits can disrupt every application relying on a provider.

With Fabric: Use health-aware routing, retries, circuit breakers and automatic fallbacks.

Inconsistent AI policies

Application-level guardrails and permissions drift as more teams and workloads are added.

With Fabric: Define policy centrally and enforce it across models, agents, tools and APIs.

Uncontrolled AI spend

Organizations struggle to understand which teams, applications and models are driving token usage and cost.

With Fabric: Attribute usage, set budgets and enforce token-aware limits in real time.

Missing governance and audit

Security teams cannot reliably determine who initiated an action, which policies applied or what an agent accessed.

With Fabric: Capture identity, routing, policy and authorization outcomes for every governed request.

How Fabric connects your AI ecosystem

Fabric separates centralized configuration and governance from runtime traffic. The control plane manages policy and identity, while the data plane routes requests inside the selected deployment environment.

Your stack

Applications

HTTP or any LLM SDK

AI agents

MCP · A2A

CLI tools

Claude Code · Codex · Cursor

Developers

Any HTTP client

Control plane

Postman Cloud

Registry

APIs · MCP · Agents

LLMs · CLIs

Policy & config manager

Versioning & syncing

Identity & RBAC

Organization → namespace

→ consumer

Admin portal

Analytics · Audit

UI & management

Data plane

Postman cloud (Free) or your cloud (Enterprise)

Protocol adapters

LLM · MCP · A2A

Plugin chain

Auth · Rate limit

PII · Transforms + more

Credential broker

OAuth · AES-256

Vault · Key management

Router

Radix tree

Weighted LB · Failover

OTel

Traces · Metrics · Logs

or export to your stack

Providers

LLM providers

1,200+ models across

70+ providers

MCP servers

REST APIs

Any HTTP backend

External & internal agents

Platform checklist

Auto-failover

Circuit breaker & retry

Rate limits

Request and token limits

Observability

Cost, latency, audit logs

Secret isolation

On managed routes, agents never see keys

Guardrails

PII redaction, access control, audit trail

Capabilities

Six capabilities. One control plane.

Built for developers and designed for the enterprise.

Universal connectivity

Connect every AI service

Route models, MCP servers, agents and internal APIs through one endpoint, with consistent authentication, governance and observability across every connection. Import a REST API from its OpenAPI spec and Fabric validates requests against it.

Policy-driven governance

Enforce policy on every call

Apply authentication, authorization, budgets and data controls centrally across teams, agents, models, tools and environments.

Credential brokering

Let agents act without exposing secrets

Agents use scoped Fabric tokens while the data plane retrieves and injects the correct upstream credential at request time.

AI-native workload support

Control AI-specific traffic

Fabric understands models, tokens, streaming responses and tool calls, enabling token-aware limits, resilient routing and configurable provider fallbacks, including on any non-2xx response, with analytics on every fallback.

Full-stack observability

Trace complete agent workflows

Follow model, MCP, agent and API calls across one workflow, with identity, policy, latency, token and cost context.

Zero-trust agent security

Verify every agent action

Validate the initiating user and executing agent, restrict delegated permissions and deny requests when identity or policy cannot be verified.

AI spend

Spend less on AI, and know where it goes

Two sides of the same bill: make every request cheaper, and see and cap what every team, user and agent spends.

Reduce AI cost

Pay less for every request, without changing your code.

  • Adaptive Model RoutingSend each prompt to the cheapest model tier that can handle it.
  • LightyearCut the tokens agent loops resend, and see the savings in analytics.
  • MCP Tool SearchKeep hundreds of tool definitions out of the context window.

Control AI spend

Know who is spending what, and stop it before the bill does.

  • Spend by team, user and agentTokens and cost attributed to whoever made the request.
  • Budgets and token limitsToken-aware limits per team, route or consumer, enforced in real time.
  • KillswitchStop traffic to a runaway agent, route or tool immediately.
Explore Cost Controls

Coverage

One gateway for every AI surface

LLMs, MCP servers and your own services, registered once and governed the same way.

Ecosystem

Works with your coding agents

Point your coding agent at Fabric and its model and MCP traffic gets the same routing, limits and traces as the rest of your stack.

Core plugins for every kind of traffic

LLMs

  • Adaptive Model RoutingPick a cheap, medium or strong model per request.
  • LightyearCut the tokens agent loops resend on every turn.
  • Provider fallbackFail over across providers, with analytics.
  • Token limitsRequest and token-based limits on LLM traffic.
  • PII redactionRedact personal data in prompts and responses.

MCPs

  • MCP Tool SearchLoad tool schemas only when the agent needs them.
  • Agent scope gateAllow only the tools an agent’s scopes permit.
  • Agent auditAn audit record for every agent action.
  • MCP HTTP bridgeServe a REST API as MCP tools from its OpenAPI spec.

APIs

  • Key and JWT authAuthenticate callers with API keys or JWTs.
  • Rate limitsCap requests per consumer, route or team.
  • ValidatorValidate requests and responses against rules.
  • TransformerRewrite requests and responses in flight.
  • Secret injectorInject upstream credentials at request time.
See all plugins in the docs

Zero Trust

Every agent action traces back to a human

When work moves between agents and tools, Fabric carries only the identity and permissions each hop needs, and denies the request when it cannot verify either.

Verify both identities

Check the initiating user and the executing agent on every request before anything runs.

Limit every delegation

Issue purpose-scoped access per hop instead of forwarding the original user token.

Authorize tool access

Allow only the tools and actions the agent is permitted to use.

Deny by default

If identity or policy cannot be verified, Fabric denies the request.

For the Enterprise

Deploy it the way your security review needs

Choose the deployment model that fits your security review.

Postman-hosted

Let Postman operate it for you with regional data planes.

Hybrid

Control plane managed by Postman; data plane stays in your cloud.

Namespaces

Give each team its own connections, services, analytics and traces, with Super admin and Member roles set per namespace.

Single sign-on

Sign in to the Fabric console through the identity provider you already use.

Bring your own provider keys

Use your own keys for each model provider; Fabric stores and injects them so apps never hold them.

Bring every AI interaction under control

Connect your first model, MCP server or internal API and apply consistent governance from one gateway.

Postman Agent Stack

Trusted by teams building on the Postman platform

Microsoft
Meta
AWS
NVIDIA
ElevenLabs
Cover Genius
Mercedes-Benz
Intuit
PayPal
HubSpot
Stripe
Uber
⌂ 换网址