How to Create a Grounded Bing Search Agent in Azure AI Foundry and Expose It as an API Endpoint

Knowledge Articles
, , , ,
1

:blue_book: Introduction

This article explains how to create a Grounded Search Agent in Azure AI Foundry using Bing Search as the external grounding tool.
Once created, the agent can be exposed as a REST API endpoint and integrated into applications such as:

  • AI for Work
  • Copilot
  • Web or backend applications
  • Slack/Teams bots
  • Enterprise services
  • Partner solutions

The agent uses Azure OpenAI models (GPT-4o, GPT-4.1, GPT-5) to process user queries and provide accurate, citation-backed, search-grounded responses.

:building_construction: Architecture Overview

+---------------------------+
|   Azure AI Foundry        |
|    Agent Runtime          |
+------------+--------------+
             |
        Tool Call
             |
             v
+---------------------------+
|    Bing Search v7 API     |
|    (Grounding Source)     |
+---------------------------+
             ^
             |
        Search Results
             |
             v
+---------------------------+
|   Azure OpenAI LLM        |
| (GPT-4o / GPT-5)          |
| Reasoning + Synthesis     |
+---------------------------+
             |
     Grounded Answer
             |
             v
+---------------------------+
|   REST Agent Endpoint     |
+---------------------------+

:jigsaw: Prerequisites

You will need:

Required Azure Resources

Resource Purpose
Azure Subscription Access to Azure AI Foundry and OpenAI
Azure AI Foundry Hub + Project Where the agent is hosted
Azure OpenAI Model Deployment GPT-4o / GPT-4.1 / GPT-5
Bing Search v7 Cognitive Service External grounding via web search

Values to Keep Handy

  • BING_SEARCH_KEY
  • BING_SEARCH_ENDPOINT
  • Model deployment name
  • Project Endpoint + API Keys

:heavy_check_mark: Step 1 — Deploy an Azure OpenAI Model

  1. Go to Azure AI Foundry → Your Project → Models

  2. Choose and deploy a model:

    • GPT-4o
    • GPT-4.1
    • GPT-5-mini

  3. Deployment Options:

    • Serverless API (recommended)
    • Dedicated deployment (for VNet or private setups)

:wrench: Step 2 — Add Bing Search as a Tool

Azure AI Foundry supports built-in tools or custom tools.

Option A — Built-in Bing Web Search Tool (Recommended)

Navigate:
AI Foundry → Agents → Tools → Add Tool → “Web Search (Bing)”

Enter your:

  • API Key
  • Endpoint
  • Region

Option B — Custom HTTP Tool (Manual Setup)

Tool Name: bing_web_search

Method: GET
Endpoint:

https://api.bing.microsoft.com/v7.0/search?q={query}

Headers:

Ocp-Apim-Subscription-Key: {{BING_SEARCH_KEY}}

Request Schema:

{
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "description": "Search query to execute on Bing"
    }
  },
  "required": ["query"]
}

:robot: Step 3 — Create the Search Agent

Navigate to:
AI Foundry → Agents → Create Agent

Agent Configuration

Setting Value
Name grounded-bing-search-agent
Runtime Model GPT-4o or GPT-5
Mode Standard Agent
Tools Attach → bing_web_search

Recommended System Prompt

You are a grounded search agent.
For all factual questions, you MUST call the bing_web_search tool first.
Use the returned web results to synthesize short, citation-backed responses.
Never hallucinate or invent facts not supported by search results.
Always cite sources or URLs in the final answer.
If a query appears subjective, still attempt grounding via search.

:test_tube: Step 4 — Test the Agent

Example query:
“Latest updates on AI copyright policies in 2025”

Expected behavior:

  1. The model calls the Bing Search tool
  2. Search results are returned
  3. The model synthesizes a grounded answer with citations

:globe_with_meridians: Step 5 — Expose the Agent as a REST API Endpoint

Azure Foundry automatically generates an endpoint.

Navigate to:
Agent → Manage → Endpoints

You will get:

Endpoint URL

https://<project>.models.ai.azure.com/agents/<agent-id>/invocations

Required Headers

Content-Type: application/json
api-key: <YOUR_PROJECT_API_KEY>

:satellite: Step 6 — How to Invoke the API

6.1. cURL Example

curl -X POST \
"https://<project>.models.ai.azure.com/agents/<agent-id>/invocations" \
-H "Content-Type: application/json" \
-H "api-key: <YOUR_API_KEY>" \
-d '{
  "messages": [
    { "role": "user", "content": "Top breakthroughs in Retrieval-Augmented Generation 2025" }
  ]
}'

6.2. JavaScript (Axios)

import axios from "axios";

const endpoint = "https://<project>.models.ai.azure.com/agents/<agent-id>/invocations";
const apiKey = "<YOUR_API_KEY>";

async function invokeAgent() {
  const response = await axios.post(
    endpoint,
    {
      messages: [
        { role: "user", content: "Summaries of latest LLM safety regulation developments" }
      ]
    },
    {
      headers: {
        "Content-Type": "application/json",
        "api-key": apiKey
      }
    }
  );

  console.log(response.data);
}

invokeAgent();

6.3. Python

import requests

endpoint = "https://<project>.models.ai.azure.com/agents/<agent-id>/invocations"
api_key = "<YOUR_API_KEY>"

payload = {
    "messages": [
        {"role": "user", "content": "Recent updates on generative AI security benchmarks"}
    ]
}

headers = {
    "Content-Type": "application/json",
    "api-key": api_key
}

res = requests.post(endpoint, headers=headers, json=payload)
print(res.json())

:rocket: Step 7 — Optional: Publish via API Management

Use Azure API Management if:

  • You want public APIs
  • You need secure routing
  • You want logging, throttling, or tenant isolation

Key Configurations

  • Backend: Agent Endpoint
  • Auth: API Keys or OAuth
  • Rate Limits
  • Logging & monitoring

:warning: Error Handling

Error Code Cause Resolution
401 Invalid/missing API key Regenerate keys
403 VNet restrictions Use private endpoint or allowlist IP
429 Rate limit exceeded Add retry logic or increase quota
ToolError Bing Search API issues Check subscription, region, or quota

:closed_lock_with_key: Security Best Practices

  • Store all keys in Azure Key Vault
  • Use Managed Identities when possible
  • Enable monitoring via Azure Monitor
  • Wrap endpoints behind API Management for added control

:paperclip: Appendix

Example Bing Search Tool Response

{
  "webPages": {
    "totalEstimatedMatches": 12,
    "value": [
      {
        "name": "AI Safety Regulation 2025",
        "url": "https://example.com/ai-policy",
        "snippet": "Governments are accelerating..."
      }
    ]
  }
}

Example Grounded Answer

Here are the latest updates on AI regulation:

  • The EU updated the AI Act with new risk-tier rules (2025).
  • The US released draft high-risk LLM regulations for healthcare and finance.
  • Japan & Singapore jointly released frontier model safety benchmarks.
1 Like