Route an AI support agent through Fabric Gateway, a runtime control layer between agents and the models, tools, and APIs they call. You connect its model, APIs, and MCP servers, enforce a token limit, and compare traces with direct connections.
- You'll build
- A fintech support agent running through Fabric Gateway with a model policy, analytics, and traces you can compare against direct connections.
- Time
- About 1 hr 15 min · 6 parts · 17 steps · progress saves as you go
- How it works
- You do each step, then click Validate and LiftOff checks your work. 11 of the 17 steps can't be checked automatically, so you click Done for those.
- You'll need
- A free Postman account and API key (connect it below)
- A few setup steps, shown below before you start
Connect a Postman API key to check your work
You do each step in Postman, then LiftOff uses your key to look at your workspace and confirm it's done. Connect once and it works for every module. Your key is saved in this browser and is only used to check your work in Postman.
What you'll practice
The supplied support agent investigates customer issues using three business systems: Banking provides accounts and transaction records, Support provides cases and evidence, and Fraud assesses transaction risk. You will:
- Explore how the agent uses these systems and an AI model to resolve a customer request.
- Run the same support agent with and without an Agent Gateway and compare their behavior.
- Configure the LLM and expose the Fraud risk-assessment API through Fabric Gateway.
- Route multiple MCP servers through Fabric Gateway and expose them through one combined route.
- Enforce a progressive tool-disclosure policy that replaces 103 MCP tool definitions with three discovery tools, reducing the Fabric lane's initial tool context from 104 tools to four.
- Enforce an LLM token limit and confirm that Fabric returns a governed 429 response.
- Configure upstream credentials and see how an Agent Gateway centralizes identity and access.
- Investigate usage, errors, and policy decisions with Analytics and Traces.
Before you begin: Connect your Postman API key in LiftOff Settings. In Part 2, you will use it to create your own workshop workspace and import three API collections, their OpenAPI specifications, and two environments. Keep your key private; enter it only in Settings, not in a collection or workshop form. The MCP collection will be forked separately.
Explore the Fintech Support Agent Architecture
In this project, you are responsible for a support agent used by a fictional fintech company. Customers contact the support team when they do not recognize a transaction, need an investigation update, or want to understand what happens next. Answering those questions safely requires information from several systems—not just a model response.
Prepare the Fintech Support Agent
Clone the support agent application and use the prepared workshop configuration to connect its Direct lane to the deployed fintech services. Then test every upstream in Postman before adding it to Fabric.
Configure the LLM and Fraud API
Fabric Gateway sits between an application or AI agent and the models, MCP servers, APIs, CLI tools, and other agents it needs to use.
Configure the MCP Servers
With the model and Fraud API now routed through Fabric, connect the Banking and Support MCP servers. You will add and test each service independently, expose both through one MCP route, and then replace the agent's large initial catalogue of 104 tools with a governed Tool Search experience. Tool Search exposes three discovery tools that progressively disclose and execute the Banking and Support tools the agent actually needs.
Run, Govern, and Observe the Agent
Compare the optimized Fabric execution path with Direct, enforce a model policy without changing the support agent, inspect the operational evidence Fabric records, and explore adaptive routing across model configurations.
Conclusion and Overview
Review the governed support-agent architecture you assembled, the differences you observed between Direct and Fabric, and the practices to carry into a production implementation.