Claude in Microsoft Foundry on Azure: Developer Guide 2026

Official Sources#

As of June 29, 2026, Claude Opus 4.8 and Claude Haiku 4.5 are generally available in Microsoft Foundry on Azure. This is the first time enterprise teams can access Claude with full Azure-native billing, identity management, and governance - no separate Anthropic contract required for initial deployment.

The practical implication: if your organization already runs on Azure with an Enterprise Agreement, you can start using Claude today with charges appearing on your existing Azure invoice. Microsoft Azure Consumption Commitment (MACC) applies.

Last updated: June 30, 2026

What Launched#

Microsoft Foundry is Azure's unified AI platform - the successor to Azure AI Studio. On June 29, Anthropic announced general availability of Claude models in Foundry with two hosting options:

Hosted on Azure (GA) - Claude runs on Anthropic-operated infrastructure within Azure data centers. Prompts and completions stay within Azure; only usage metadata and safety-flagged content egress to Anthropic. Available models: Claude Opus 4.8 and Claude Haiku 4.5.

Hosted on Anthropic (Preview) - Claude runs on Anthropic's own infrastructure. Supports all Claude models including Fable 5 and the full Opus/Sonnet lineup. Use this for features not yet available on Azure-hosted deployments.

The Hosted on Azure option runs on NVIDIA GB300 Blackwell Ultra systems, which Microsoft positions for autonomous and domain-specific AI agents.

Model Availability#

All models on Foundry have 1M-token context windows except Claude Sonnet 4.5, which has 200k.

CCU Pricing#

Microsoft Foundry bills through Azure Marketplace using Claude Consumption Units (CCUs). The conversion is straightforward: 100 CCU = $1.00 USD of Claude usage at standard Anthropic API rates.

The underlying token pricing matches Anthropic's standard API rates:

Prompt caching multipliers apply: 5-minute cache writes cost 1.25x input, 1-hour cache writes cost 2x input, cache hits cost 0.1x input. Batch API discounts (50% off) are available for async workloads.

US Data Zone pricing: Using the US Data Zone Standard deployment type adds a 1.1x multiplier to all token pricing. This keeps inference within the United States.

Getting Started#

Prerequisites#

• An active Azure subscription
• Access to Microsoft Foundry
• Azure CLI installed (optional, for resource management)

Step 1: Create a Foundry Resource#

1. Navigate to the Foundry portal
2. Create a new Foundry resource or select an existing one
3. Configure access management using Azure-issued API keys or Entra ID
4. Optionally configure private network (Azure Virtual Network)
5. Note your resource name - you will use this as {resource} in API endpoints

Your endpoint URL format: https://{resource}.services.ai.azure.com/anthropic/v1/*

Step 2: Deploy a Claude Model#

1. In the Foundry portal, select Discover > Models
2. Search for a Claude model (e.g., claude-opus-4-8)
3. Select Deploy > Custom settings
4. On your first Claude deployment, accept the Azure Marketplace terms
5. Configure the deployment:
   • Deployment name: Defaults to model ID, but you can customize
   • Region scope: Global or Data Zone (US only)
   • Model version: Select hosting option (Hosted on Azure or Hosted on Anthropic)
6. Select Deploy and wait for provisioning

Step 3: Get Your Credentials#

1. In Foundry, select Build > Models
2. Open your Claude deployment and select the Details tab
3. Copy the Key and note the Target URI

SDK Installation#

Foundry is supported by Python, TypeScript, C#, Java, and PHP SDKs.

Python:

pip install -U "anthropic"

TypeScript:

npm install @anthropic-ai/foundry-sdk

C#:

dotnet add package Anthropic.Foundry

Java (Gradle):

implementation("com.anthropic:anthropic-java-foundry:2.40.0")

Authentication Options#

API Key Authentication#

Set environment variables:

export ANTHROPIC_FOUNDRY_API_KEY="your-api-key"
export ANTHROPIC_FOUNDRY_RESOURCE="your-resource-name"

Python example:

import os
from anthropic import AnthropicFoundry

client = AnthropicFoundry(
    api_key=os.environ.get("ANTHROPIC_FOUNDRY_API_KEY"),
    resource="example-resource",
)

message = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(message.content)

TypeScript example:

import AnthropicFoundry from "@anthropic-ai/foundry-sdk";

const client = new AnthropicFoundry({
  apiKey: process.env.ANTHROPIC_FOUNDRY_API_KEY,
  resource: "example-resource"
});

const message = await client.messages.create({
  model: "claude-opus-4-8",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello!" }]
});
console.log(message.content);

cURL example:

curl https://{resource}.services.ai.azure.com/anthropic/v1/messages \
  -H "content-type: application/json" \
  -H "api-key: YOUR_AZURE_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Entra ID Authentication#

For enterprise deployments, use Microsoft Entra ID (formerly Azure Active Directory) for centralized access management via Azure RBAC.

Python with Entra ID:

import os
from anthropic import AnthropicFoundry
from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(), "https://ai.azure.com/.default"
)

client = AnthropicFoundry(
    resource="example-resource",
    azure_ad_token_provider=token_provider,
)

message = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(message.content)

cURL with Entra ID:

ACCESS_TOKEN=$(az account get-access-token --resource https://ai.azure.com --query accessToken -o tsv)

curl https://{resource}.services.ai.azure.com/anthropic/v1/messages \
  -H "content-type: application/json" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Feature Limitations on Azure#

When using Hosted on Azure deployments, the following features are not available:

• Structured outputs
• Server-side tools (web search, web fetch, code execution, tool search)
• MCP connector
• Agent Skills
• Programmatic tool calling
• Files API

Requests that use these features return a 400 Bad Request error. Claude Code automatically detects Hosted on Azure deployments and adapts its feature set.

For full feature parity, use Hosted on Anthropic deployments instead.

Monitoring and Logging#

Azure provides native observability for Claude usage:

• Azure Monitor: Track API usage, latency, and error rates
• Log Analytics: Query and analyze request/response logs
• Cost Management: Monitor and forecast CCU consumption

For debugging, include both request-id and apim-request-id response headers when contacting support.

Migration Between Hosting Options#

To move from Hosted on Anthropic to Hosted on Azure (or vice versa):

1. Create a new deployment with the other hosting version
2. Update your application to pass the new deployment name in the model parameter
3. Delete the old deployment once traffic has moved

If the new deployment is in the same Foundry resource, your endpoint URL and authentication stay unchanged.

California Government Partnership#

On June 29, Governor Newsom announced that California signed a first-of-its-kind agreement with Anthropic giving state agencies, cities, and counties access to Claude at a 50% discount. The deal includes free workforce training, technical assistance, and workflow help from Anthropic developers.

This partnership runs through Azure Marketplace, using the same CCU billing structure. Public sector organizations in California should contact their Anthropic or Microsoft account representative for access.

When to Use Foundry vs Direct API#

Use Microsoft Foundry when:

• Your organization is already on Azure with an Enterprise Agreement
• You need MACC credits to apply to AI spend
• Centralized Azure billing and governance matter
• Entra ID for identity management is a requirement
• You want Claude alongside other Foundry models in one platform

Use the direct Anthropic API when:

• You need Fable 5, Claude Managed Agents, or Batch API features
• Server-side tools (web search, code execution) are required
• You want the full feature set without hosting limitations
• Multi-cloud or cloud-agnostic deployment is important

FAQ#

How does Microsoft Foundry billing compare to the Anthropic API?#

The underlying token rates are identical. Foundry adds no markup - you pay standard Anthropic rates, converted to CCUs for Azure billing. The main difference is payment method (Azure invoice vs Anthropic billing) and MACC eligibility.

Can I use Claude Code with Foundry deployments?#

Yes. Claude Code detects Hosted on Azure deployments automatically and adapts its feature set. Some features like server-side tools are not available, but core coding assistance works.

What is a Claude Consumption Unit (CCU)?#

A CCU is a billing unit for Azure Marketplace. 100 CCU = $1.00 USD of Claude usage at standard rates. CCUs are metered hourly and invoiced monthly in arrears.

Are prompt caching and batch processing available on Foundry?#

Prompt caching is available with the same multipliers as the direct API. Batch API with 50% discount is available. Fast mode is not available on Claude Platform on AWS but should be checked for Foundry availability.

What models are available on Hosted on Azure?#

Currently, Claude Opus 4.8 and Claude Haiku 4.5 are available on Hosted on Azure. Other models including Fable 5 and the Sonnet family are available on Hosted on Anthropic (preview).

How do I get US data residency on Foundry?#

Use the US Data Zone Standard deployment type when creating your deployment. This keeps inference within the United States and applies a 1.1x pricing multiplier.

Can I migrate existing Anthropic API applications to Foundry?#

Yes. The SDK changes are minimal - swap Anthropic for AnthropicFoundry and provide your resource name. API request/response formats are identical.

What support options are available?#

Include both request-id and apim-request-id response headers when contacting support to help teams locate your request across both Anthropic and Azure systems.

Continue Reading#

• Claude Code Fast Mode: When 2.5x Speed Is Worth 2x Price
• Cursor Composer 2.5 Developer Guide 2026
• GitHub Copilot's Impact Dashboard Now Puts a Dollar Figure on Agent-First Development
• MiniMax M2.5 for Developers: The Anthropic-Compatible Budget Frontier Model

Sources#

• Claude in Microsoft Foundry documentation - verified June 30, 2026
• Claude in Microsoft Foundry pricing - verified June 30, 2026
• Azure blog: Claude in Microsoft Foundry GA - June 29, 2026
• Anthropic: Claude in Microsoft Foundry announcement - June 29, 2026
• California Government Claude partnership - June 29, 2026
• Microsoft Learn: Deploy Claude in Foundry - June 2026