MCP Quickstart

Set up the Proxy MCP server with Claude, Cursor, and other AI tools

MCP Quickstart

This guide walks you through connecting the Proxy MCP server to your preferred AI assistant.

Prerequisites

• A Proxy account with completed KYC (sign up)
• An AI assistant that supports MCP (Claude Desktop, Cursor, etc.)

Server URL

The Proxy MCP server is hosted at:

https://mcp.useproxy.ai/api/mcp

Authentication

The MCP server uses OAuth 2.0 for authentication. You’ll need to authorize the connection through your Proxy account.

The OAuth flow will redirect you to Proxy’s authentication page. After signing in, the connection will be established automatically.

Setup by Platform

Claude Desktop

Claude Desktop Setup

1. Open Claude Desktop settings
2. Navigate to Developer > Model Context Protocol
3. Click Add Server
4. Configure the server:

{
  "mcpServers": {
    "proxy": {
      "url": "https://mcp.useproxy.ai/api/mcp",
      "transport": "streamable-http"
    }
  }
}

5. Click Save and restart Claude Desktop
6. When prompted, authorize the connection through Proxy

Verify Connection

Ask Claude: “What tools do you have from Proxy?”

Claude should list the available Proxy tools including agent.create, card.list, etc.

Cursor

Cursor Setup

1. Open Cursor Settings (Cmd/Ctrl + ,)
2. Navigate to Features > MCP Servers
3. Click Add new MCP server
4. Enter the configuration:

Name: proxy URL: https://mcp.useproxy.ai/api/mcp Transport: streamable-http

5. Save and authorize when prompted

Using in Cursor

In the Composer or Chat, you can now ask Cursor to interact with Proxy:

Create a new agent called "Code Assistant" for handling software purchases

Custom Client

Custom MCP Client

If you’re building a custom MCP client, connect using these parameters:

Endpoint: https://mcp.useproxy.ai/api/mcp Transport: HTTP with Server-Sent Events (SSE) Authentication: OAuth 2.0 Bearer token

OAuth Configuration

Discover OAuth metadata at:

https://mcp.useproxy.ai/.well-known/oauth-protected-resource

Example Connection (Node.js)

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@anthropic-ai/mcp-client";

const transport = new StreamableHTTPClientTransport({
  url: "https://mcp.useproxy.ai/api/mcp",
  headers: {
    Authorization: `Bearer ${accessToken}`,
  },
});

const client = new Client({
  name: "my-app",
  version: "1.0.0",
});

await client.connect(transport);

// List available tools
const tools = await client.listTools();
console.log(tools);

// Call a tool
const result = await client.callTool("agent.list", {});
console.log(result);

JSON-RPC Protocol

The server uses JSON-RPC 2.0. Direct HTTP calls:

# List tools
curl -X POST https://mcp.useproxy.ai/api/mcp \
  -H "Authorization: Bearer your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/list"
  }'

# Call a tool
curl -X POST https://mcp.useproxy.ai/api/mcp \
  -H "Authorization: Bearer your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "agent.list",
      "arguments": {}
    }
  }'

First Steps

Once connected, try these example commands:

Check Your Account Status

What's my onboarding status?

Create an Agent

Create an agent called "Shopping Assistant" for handling e-commerce purchases

Issue a Card

Create a virtual card for my Shopping Assistant agent with a $200 daily limit

Check Balance

What's my current balance?

Agent Authentication

For autonomous agent workflows, you can authenticate using an agent token instead of user OAuth.

Generate an Agent Token

1. In your AI assistant, ask:

   Create a token for my Shopping Assistant agent
   

2. Save the returned token securely (it won’t be shown again)

Use Agent Token

Agent tokens can be used directly in the Authorization header:

curl -X POST https://mcp.useproxy.ai/api/mcp \
  -H "Authorization: Bearer agt_xxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "intent.create",
      "arguments": {
        "cardId": "card_xxx",
        "summary": "Purchase development tools",
        "expectedAmount": 9900
      }
    }
  }'

Agent tokens have limited permissions. They can only access intent, access, card (read-only), transaction (read-only), and balance tools.

Troubleshooting

Connection refused or timeout

• Verify the server URL is correct: https://mcp.useproxy.ai/api/mcp
• Check that your network allows outbound HTTPS connections
• Try the health endpoint: curl https://mcp.useproxy.ai/api/mcp/health

Authentication failed

• Ensure you completed the OAuth flow
• Try disconnecting and reconnecting the MCP server
• Check that your Proxy account is active

Tool not found

• Verify the tool name matches exactly (e.g., agent.create not createAgent)
• Ensure you’re using the correct authentication type (some tools require user auth, others agent auth)

Permission denied

• Check that you own the resource you’re trying to access
• Agent tokens can only access their own cards and intents
• Some operations require user authentication (not agent tokens)

Next Steps

Tools Reference

Complete documentation of all 29 MCP tools

Agents

Learn about agent architecture