OpenAI

1
{
2
  "mcpServers": {
3
    "respan-docs": {
4
      "url": "https://mcp.respan.ai/mcp/docs"
5
    }
6
  }
7
}

Use Respan Gateway to call OpenAI models while keeping unified observability (logs, cost, latency, and reliability metrics) in Respan.

Prerequisites

• A Respan API key
• An OpenAI API key (BYOK)

Supported SDKs / integrations

Configuration

There are 2 ways to add your OpenAI credentials to your requests:

Via UI (Global)

1

Navigate to Providers

Go to the Providers page. This page allows you to manage credentials for over 20+ supported providers.

2

Add your OpenAI API Key

Select OpenAI and paste your API key.

3

Configure Load Balancing (Optional)

You can add multiple OpenAI API keys for redundancy. Use the Load balancing weight field to determine how traffic is distributed between keys.

Via code (Per-Request)

You can pass credentials dynamically in the request body. This is useful if you need to use your users’ own API keys (BYOK).

Add the customer_credentials parameter to your Gateway request:

1
{
2
  // Rest of the request body
3
  "customer_credentials": {
4
    "openai": {
5
      "api_key": "YOUR_OPENAI_API_KEY",
6
    }
7
  }
8
}

Log OpenAI requests

If you are not using the Gateway to proxy requests, you can still log your OpenAI requests to Respan asynchronously. This allows you to track cost, latency, and performance metrics for external calls.

1
import requests
2
3
url = "https://api.respan.ai/api/request-logs/create/"
4
payload = {
5
    "model": "gpt-4o",
6
    "prompt_messages": [
7
        {
8
            "role": "user",
9
            "content": "Hello, how are you?"
10
        }
11
    ],
12
    "completion_message": {
13
        "role": "assistant",
14
        "content": "I'm doing well, thank you for asking!"
15
    },
16
    "cost": 0.0015,
17
    "generation_time": 2.3,
18
    "customer_params": {
19
        "customer_identifier": "user_123"
20
    }
21
}
22
headers = {
23
    "Authorization": "Bearer YOUR_RESPAN_API_KEY",
24
    "Content-Type": "application/json"
25
}
26
27
response = requests.post(url, headers=headers, json=payload)