Autogenerate MCP tools from a RESTful API

Before using the AI MCP Proxy plugin, you’ll need an upstream HTTP API to expose. For this tutorial, we’ll use a simple mock API built with Express. This allows you to test the plugin without relying on an external service. This mock API simulates a small marketplace system with a fixed set of users and their associated orders. Each user has between two and five sample orders, which the API exposes through /marketplace/users and /marketplace/{userId}/orders endpoints.

Running these commands will download the mock API script and install any required dependencies automatically:

curl -s -o api.js "https://gist.githubusercontent.com/subnetmarco/5ddb23876f9ce7165df17f9216f75cce/raw/a44a947d69e6f597465050cc595b6abf4db2fbea/api.js"
npm install express
node api.js

Validate the API is running:

curl -X GET http://localhost:3000

This request confirms that the mock server is up and responding. Later, the AI MCP Proxy will use this API’s OpenAPI schema to generate MCP tool definitions. You should see the following response from the server:

{"name":"Sample Users API"}%

With the mock API server running, configure the AI MCP Proxy plugin to expose its endpoints as MCP tools. The following example maps the mock API operations to MCP tool definitions that the client can invoke.

echo '
_format_version: "3.0"
plugins:
  - name: ai-mcp-proxy
    route: mcp-route
    config:
      mode: conversion-listener
      tools:
      - description: Get users
        method: GET
        path: "/marketplace/users"
        parameters:
        - name: id
          in: query
          required: false
          schema:
            type: string
          description: Optional user ID
      - description: Get orders for a user
        method: GET
        path: "/marketplace/orders"
        parameters:
        - description: User ID to filter orders
          in: query
          name: userid
          required: true
          schema:
            type: string
      server:
        timeout: 60000
' | deck gateway apply -

1. Open your Cursor desktop app.

2. Navigate to Cursor > Settings.

3. In the Settings tab, go to Tools & integrations in the left sidebar.

4. In the MCP Tools section, click Add Custom MCP.

5. Paste the following JSON configuration into the newly opened mcp.json tab:

   {
     "mcpServers": {
       "marketplace": {
         "url": "http://localhost:8000/marketplace"
       }
     }
   }
   

   Copied!

6. Return to the Cursor settings tab. You should now see the Marketplace MCP server with two tools available:

   

7. To open a new Cursor chat, click cmd + L if you’re on Mac, or ctrl + L if you’re on Windows.

8. In the Cursor chat tab, click @ Add Context and select mcp.json:

Enter the following question in the Cursor chat:

What users do you see in the API?

I'll query the marketplace service for the list of users
> Called mcp-route-1

When the agent finishes reasoning, You should see the following output:

Let me check what users are available in the API:
I can see 10 users in the API:
1. Alice Johnson (ID: a1b2c3d4)
2. Bob Smith (ID: e5f6g7h8)
3. Charlie Lee (ID: i9j0k1l2)
4. Diana Evans (ID: m3n4o5p6)
5. Ethan Brown (ID: q7r8s9t0)
6. Fiona Clark (ID: u1v2w3x4)
7. George Harris (ID: y5z6a7b8)
8. Hannah Lewis (ID: c9d0e1f2)
9. Ian Walker (ID: g3h4i5j6)
10. Julia Turner (ID: k7l8m9n0)

Now, we can check what Alice Johnson ordered by entering the following message in the Cursor chat:

What did Alice Johnson order?

You’ll notice that Cursor calls the tools exposed by the AI MCP Proxy plugin:

I'll look up the list of users to find Alice's user ID. Then I'll fetch her orders.

> called mcp-route-1
> called mcp-route-2

When the agent finishes reasoning, you should see the following response:

Sugar (50kg)
Cleaning Supplies Pack
Canned Tomatoes (100 cans)

You can validate this result against the API exposed in the previous step.