Foundry IQ Python SDK

Build agentic retrieval pipelines using Azure AI Search knowledge bases with the Python SDK.

Installation

pip install azure-ai-projects==2.0.0b1 azure-search-documents==11.7.0b2 azure-identity

Environment Variables

AZURE_SEARCH_ENDPOINT=https://<search-service>.search.windows.net
AZURE_OPENAI_ENDPOINT=https://<openai-resource>.openai.azure.com
AZURE_OPENAI_EMBEDDING_DEPLOYMENT=text-embedding-3-large
AZURE_AI_PROJECT_ENDPOINT=https://<resource>.services.ai.azure.com/api/projects/<project>

Authentication

from azure.identity import DefaultAzureCredential
from azure.search.documents.indexes import SearchIndexClient

credential = DefaultAzureCredential()
index_client = SearchIndexClient(endpoint=search_endpoint, credential=credential)

Architecture

User Query → Foundry Agent → MCP Tool → Knowledge Base → Knowledge Sources
                                              ↓
                              Query Planning + Hybrid Search + Reranking
                                              ↓
                              Extractive Data with Citations

Core Workflow

1. Create Search Index (Semantic Config Required)

from azure.search.documents.indexes import SearchIndexClient
from azure.search.documents.indexes.models import (
    SearchIndex, SearchField, VectorSearch, VectorSearchProfile,
    HnswAlgorithmConfiguration, AzureOpenAIVectorizer,
    AzureOpenAIVectorizerParameters, SemanticSearch,
    SemanticConfiguration, SemanticPrioritizedFields, SemanticField
)

index = SearchIndex(
    name=index_name,
    fields=[
        SearchField(name="id", type="Edm.String", key=True, filterable=True),
        SearchField(name="content", type="Edm.String", searchable=True),
        SearchField(name="embedding", type="Collection(Edm.Single)",
                   stored=False, vector_search_dimensions=3072,
                   vector_search_profile_name="hnsw-profile"),
    ],
    vector_search=VectorSearch(
        profiles=[VectorSearchProfile(name="hnsw-profile", algorithm_configuration_name="hnsw-algo", vectorizer_name="aoai-vectorizer")],
        algorithms=[HnswAlgorithmConfiguration(name="hnsw-algo")],
        vectorizers=[AzureOpenAIVectorizer(
            vectorizer_name="aoai-vectorizer",
            parameters=AzureOpenAIVectorizerParameters(
                resource_url=aoai_endpoint,
                deployment_name="text-embedding-3-large",
                model_name="text-embedding-3-large"
            )
        )]
    ),
    semantic_search=SemanticSearch(  # REQUIRED for agentic retrieval
        default_configuration_name="semantic-config",
        configurations=[SemanticConfiguration(
            name="semantic-config",
            prioritized_fields=SemanticPrioritizedFields(
                content_fields=[SemanticField(field_name="content")]
            )
        )]
    )
)
index_client.create_or_update_index(index)

2. Create Knowledge Source

from azure.search.documents.indexes.models import (
    SearchIndexKnowledgeSource, SearchIndexKnowledgeSourceParameters, SearchIndexFieldReference
)

ks = SearchIndexKnowledgeSource(
    name="my-knowledge-source",
    description="Knowledge source for retrieval",
    search_index_parameters=SearchIndexKnowledgeSourceParameters(
        search_index_name=index_name,
        source_data_fields=[SearchIndexFieldReference(name="id"), SearchIndexFieldReference(name="content")]
    )
)
index_client.create_or_update_knowledge_source(knowledge_source=ks)

3. Create Knowledge Base

from azure.search.documents.indexes.models import (
    KnowledgeBase, KnowledgeBaseAzureOpenAIModel, KnowledgeSourceReference,
    AzureOpenAIVectorizerParameters, KnowledgeRetrievalOutputMode,
    KnowledgeRetrievalLowReasoningEffort
)

aoai_params = AzureOpenAIVectorizerParameters(
    resource_url=aoai_endpoint,
    deployment_name="gpt-4.1-mini",
    model_name="gpt-4.1-mini"
)

kb = KnowledgeBase(
    name="my-knowledge-base",
    knowledge_sources=[KnowledgeSourceReference(name="my-knowledge-source")],
    models=[KnowledgeBaseAzureOpenAIModel(azure_open_ai_parameters=aoai_params)],
    output_mode=KnowledgeRetrievalOutputMode.EXTRACTIVE_DATA,  # Recommended for agent integration
    retrieval_reasoning_effort=KnowledgeRetrievalLowReasoningEffort()
)
index_client.create_or_update_knowledge_base(knowledge_base=kb)

mcp_endpoint = f"{search_endpoint}/knowledgebases/{kb.name}/mcp?api-version=2025-11-01-preview"

4. Create Project Connection

import requests
from azure.identity import get_bearer_token_provider

bearer_token = get_bearer_token_provider(credential, "https://management.azure.com/.default")()

response = requests.put(
    f"https://management.azure.com{project_resource_id}/connections/{connection_name}?api-version=2025-10-01-preview",
    headers={"Authorization": f"Bearer {bearer_token}"},
    json={
        "name": connection_name,
        "type": "Microsoft.MachineLearningServices/workspaces/connections",
        "properties": {
            "authType": "ProjectManagedIdentity",
            "category": "RemoteTool",
            "target": mcp_endpoint,
            "isSharedToAll": True,
            "audience": "https://search.azure.com/",
            "metadata": {"ApiType": "Azure"}
        }
    }
)
response.raise_for_status()

5. Create Agent with MCP Tool

from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import PromptAgentDefinition, MCPTool

client = AIProjectClient(endpoint=project_endpoint, credential=credential)

instructions = """You are a helpful assistant that must use the knowledge base to answer all questions from user. You must never answer from your own knowledge under any circumstances.
Every answer must always provide annotations for using the MCP knowledge base tool and render them as: 【message_idx:search_idx†source_name】
If you cannot find the answer in the provided knowledge base you must respond with "I don't know"."""

mcp_tool = MCPTool(
    server_label="knowledge-base",
    server_url=mcp_endpoint,
    require_approval="never",
    allowed_tools=["knowledge_base_retrieve"],
    project_connection_id=connection_name
)

agent = client.agents.create_version(
    agent_name="my-agent",
    definition=PromptAgentDefinition(model="gpt-4.1-mini", instructions=instructions, tools=[mcp_tool])
)

6. Invoke Agent

openai_client = client.get_openai_client()
conversation = openai_client.conversations.create()

response = openai_client.responses.create(
    conversation=conversation.id,
    tool_choice="required",  # Ensures agent always uses knowledge base
    input="What are the key findings?",
    extra_body={"agent": {"name": agent.name, "type": "agent_reference"}}
)
print(response.output_text)

Query Knowledge Base Directly

from azure.search.documents.knowledgebases import KnowledgeBaseRetrievalClient
from azure.search.documents.knowledgebases.models import (
    KnowledgeBaseRetrievalRequest, KnowledgeBaseMessage,
    KnowledgeBaseMessageTextContent, SearchIndexKnowledgeSourceParams
)

kb_client = KnowledgeBaseRetrievalClient(endpoint=search_endpoint, knowledge_base_name="my-knowledge-base", credential=credential)

request = KnowledgeBaseRetrievalRequest(
    messages=[KnowledgeBaseMessage(role="user", content=[KnowledgeBaseMessageTextContent(text="What is vector search?")])],
    knowledge_source_params=[SearchIndexKnowledgeSourceParams(
        knowledge_source_name="my-knowledge-source",
        include_references=True,
        include_reference_source_data=True
    )],
    include_activity=True
)

result = kb_client.retrieve(request)
print(result.response[0].content[0].text)

SharePoint with User Token

For remote SharePoint sources, pass user token for ACL trimming:

from azure.identity import get_bearer_token_provider

mcp_tool = MCPTool(
    server_label="knowledge-base",
    server_url=mcp_endpoint,
    require_approval="never",
    allowed_tools=["knowledge_base_retrieve"],
    project_connection_id=connection_name,
    headers={"x-ms-query-source-authorization": get_bearer_token_provider(credential, "https://search.azure.com/.default")()}
)

Retrieval Reasoning Effort

Level	Class	Best For
Minimal	KnowledgeRetrievalMinimalReasoningEffort()	Simple lookups, lowest cost/latency
Low	KnowledgeRetrievalLowReasoningEffort()	Standard queries (default)
Medium	KnowledgeRetrievalMediumReasoningEffort()	Complex multi-hop questions

Output Modes

Mode	Value	Use Case
Extractive Data	KnowledgeRetrievalOutputMode.EXTRACTIVE_DATA	Agent integration (recommended)
Answer Synthesis	KnowledgeRetrievalOutputMode.ANSWER_SYNTHESIS	Direct KB responses with citations

Supported LLMs

gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-5, gpt-5-mini, gpt-5-nano

API Version

api-version=2025-11-01-preview

Prerequisites

• Azure AI Search with semantic ranker enabled
• Microsoft Foundry project with LLM deployment and system-assigned managed identity
• Required roles:
  • Search Service Contributor: Create objects
  • Search Index Data Reader: Read indexed content (assign to project managed identity)
  • Azure AI User: Access model deployments, create agents
  • Azure AI Project Manager: Create project connections

Reference Files

• references/patterns.md: Complete SDK patterns for all operations
• references/knowledge-sources.md: Knowledge source configurations
• scripts/create_agent.py: End-to-end script