How to Use Claude in Microsoft Copilot Studio: A Step-by-Step Guide

TL;DR: Microsoft Copilot Studio now supports Anthropic Claude models directly through the AI Model Hub. If your tenant is Claude-enabled, you can integrate Claude into your Copilot agents in minutes. We will show you exactly how to do it. 

You've built agents in Microsoft Copilot Studio that pull information from SharePoint, answer HR-related questions, and manage IT triage tasks. These agents are working well. However, you've noticed that Anthropic's Claude models are known for handling long documents, performing complex reasoning, and understanding nuanced language with remarkable accuracy. You're wondering whether you can integrate Claude into your existing Copilot Studio environment. 

The answer is yes. Multi-model AI is one of the key Microsoft 365 trends for 2026. So, if your tenant is already enabled for Claude, you can have it up and running in less than 30 minutes. 

What Actually Changed: Copilot Studio Became Multi-Model 

Here's what often slows teams down: the belief that adding a non-Microsoft model to your Copilot environment requires a new procurement process, licensing, or starting over with a new agent framework. But the good news is that's no longer the case! 

Microsoft made it easier to build AI models in Copilot Studio by removing barriers. Now, Claude models from Anthropic are easily accessible through the generative AI model selector in the Copilot Studio interface. As Microsoft's documentation explains, this feature lets users pick the model that best fits their needs without leaving the platform. 

However, there's a small catch — your tenant needs to be ready. Copilot readiness isn't just about having a license; it's about ensuring the right setup is in place so features like support for third-party models can work smoothly. If any step is missed, the option won't appear in your UI. 

Let's fix that. 

Step 1: Confirm Your Copilot Studio Licensing 

Claude support in Copilot Studio requires an active Microsoft Copilot Studio license, either the standalone Copilot Studio subscription or Microsoft 365 Copilot with Copilot Studio capacity. If you're unsure what your tenant has, check the Microsoft 365 Admin Center under Billing > Licenses. For more information, read Microsoft's documentation on selecting AI models in Copilot Studio.

You'll need at least one of the following: 

• Copilot Studio (per tenant/per user)
• Microsoft 365 Copilot with Copilot Studio included
• If you're on a trial or a legacy Power Virtual Agents license, Claude model access will not be available. 

Step 2: Enable Third-Party Model Access in Your Tenant 

This step is often overlooked in many guides. Before you can select Claude as a model in Copilot Studio, a Global Admin or Power Platform Admin must enable third-party AI model connectors in the Power Platform Admin Center. As of April 2026, admins can scope third-party model access to specific users or Entra ID groups.

Follow these steps: 

1. Navigate to the Power Platform Admin Centre at admin.powerplatform.microsoft.com.
2. Select the environment you want to configure.
3. Go to Settings > Features.
4. Under AI Builder and AI Models, make sure "Allow users to access generative AI features" is toggled on.
5. Ensure that "Allow third-party AI models" is enabled. This option may be listed under preview features depending on your tenant region. 

After making these changes, save your settings and allow up to 24 hours for the updates to take effect across your environment. 

Step 3: Open Copilot Studio and Create or Open an Agent 

Once the tenant configuration is confirmed, log in to Copilot Studio (copilotstudio.microsoft.com) using your maker account. You have two options: 

1. Create a new agent from scratch. 
2. Open an existing agent that you want to migrate to Claude. 

For testing purposes, creating a new agent is the preferred approach. It allows you to validate Claude's behaviour before committing a production agent to the new model. 

Not sure what to build? Here are five high-impact agent use cases to get started.

Step 4: Select Claude as Your Generative Model 

Here's where the real value comes in. Open your agent's Settings, then go to Generative AI. You'll find a model selector under "Which AI model should power your agent?" If your tenant is set up correctly, you'll see Anthropic Claude models alongside GPT-4o and other options. Pick the Claude model that suits your needs. Not sure which model to pick? Anthropic's guide breaks down when to use Haiku, Sonnet, or Opus.

Copilot Studio supports: 

• Claude 3 Haiku (quick, affordable, ideal for FAQs and triage)
• Claude 3 Sonnet (well-rounded for document-heavy tasks)
• Claude 3 Opus (most powerful, best for complex reasoning and multi-step workflows) 

Choose based on what you're working on. For SharePoint agents that handle long documents, Sonnet is usually the best place to start. 

Step 5: Configure Your Knowledge Sources 

Claude, similar to GPT-4o, relies on your configured knowledge sources, such as SharePoint sites, uploaded files, or manual Q&A, to base its responses. You don't need to modify your knowledge source setup when switching models. 

To connect to SharePoint content: 

1. Navigate to Knowledge in your agent 
2. Click on Add knowledge source
3. Select SharePoint
4. Enter the URL of your SharePoint site or document library
5. Save 

Your agent's quality depends on your SharePoint content. Here's how to improve findability before connecting knowledge sources. Claude will then use that content as its source for retrieval when answering questions.

Step 6: Test Before You Publish 

It's best to avoid publishing directly to production. Instead, try using the built-in test panel in Copilot Studio to run your queries against your Claude-powered agent. This way, you can specifically check:

• How it handles long-document queries (leveraging Claude's strong context window)
• Multi-turn conversations that require memory
• Edge cases where your previous model faced difficulties 

Make sure to compare these outputs with those from your previous model. Since Claude often interprets ambiguous or nuanced prompts differently, and often more accurately, it's always a good idea to validate its responses against your specific knowledge base. 

Step 7: Publish and Monitor 

Once you're happy with how your agent is performing, go ahead and publish it to your preferred platform, whether that's Teams, SharePoint, your website, or elsewhere. 

After launch, keep an eye on its progress using Copilot Studio's analytics dashboard, which shows key metrics including session volume, escalation rate, and topic coverage. The Tenant Dashboard for Microsoft 365 also provides a complete view of your agent's activity and adoption across your environment, helping you manage everything with ease. 

Copilot Readiness: Why Most 'Claude Problems' Aren't Claude Problems

None of these steps work if your tenant isn't Copilot-ready. Copilot readiness means your licensing is clean, your Power Platform environments are properly structured, your data governance policies are set, and your admin configurations are enabled before your makers start building. 

Enabling third-party models like Claude? Make sure your AI workplace policy covers it.

Organizations that rush to build agents without laying this groundwork end up troubleshooting configuration issues instead of shipping value. A structured Copilot readiness assessment closes that gap before it costs you time. If Claude doesn't appear in your model selector, the answer is almost always a missed tenant configuration step — not a missing license. 

Model Choice Is Here, but Your Readiness Determines the Outcome 

Claude in Copilot Studio isn't just a workaround; it's now an officially supported model option in a platform designed to help teams choose the best model for their needs. For SharePoint admins working with many documents, it's definitely worth giving it a try. 

You might notice improvements in how your agent handles long inputs, tricky cases, and detailed questions, but the only way to be sure is to test it against your own content and "golden set" prompts. 

The steps mentioned earlier are the technical details, but what really makes the difference between a smooth rollout and a frustrating experience is ensuring your tenant and environment are set up to support the external models you want to use. 

Once it's sorted out, take your time. Run a thorough pilot, compare results for accuracy, escalations, and consistency, and then implement the change with confidence.  

Next Step: Make Claude a Configuration Win 

Not sure whether your tenant is configured correctly to support Claude or other advanced Copilot Studio features? 

We'll run a focused Copilot Readiness Assessment to map:

• Tenant-level provider enablement (what's allowed and for whom)
• Environment controls for external models (where it's enabled and where it shouldn't be)
• Governance and rollout approach so makers can build without hitting invisible gates 

Book your Copilot Readiness Assessment now.

Claude in Copilot Studio FAQs