Does Claude Work in Microsoft Copilot Studio? (2026)

The short answer: Microsoft Copilot Studio does not list Anthropic's Claude models in its native model picker. The supported models are GPT-4o, GPT-5, and OpenAI's family via Azure OpenAI Service, plus Microsoft's own Phi family. To use Claude inside a Copilot Studio workflow, you build a custom connector that wraps the Claude API and call it through an HTTP action node.

That's the technical answer. The strategic answer is more nuanced. Anthropic and Microsoft expanded their partnership in late 2025 to bring Claude to Azure AI Foundry — but Foundry is a different product than Copilot Studio. Foundry is for engineers building custom AI applications; Copilot Studio is for ops teams building no-code agents inside Microsoft 365. The Claude-in-Microsoft path goes through Foundry, not Copilot Studio.

This guide covers what's actually possible in 2026, the custom-connector workaround, when each path is the right call, and the cost math.

The native model picker — what Copilot Studio supports out of the box

As of mid-2026, Copilot Studio's "Generative AI" model selector exposes these options:

• GPT-4o — OpenAI's multimodal flagship, via Azure OpenAI
• GPT-4.1 and GPT-4.1 mini — OpenAI's 2025-era models, still available for backward compatibility
• GPT-5 — OpenAI's reasoning-tier model
• GPT-5.4 — the balanced workhorse tier
• o3-mini — for explicit reasoning tasks
• Microsoft Phi-3 / Phi-4 — Microsoft's own family for cost-sensitive workloads

No Claude. No Gemini. No DeepSeek. No Llama. The native picker is OpenAI-and-Microsoft-only.

This is a deliberate product choice tied to Microsoft's commercial relationship with OpenAI — Microsoft is a major OpenAI investor, and Azure OpenAI Service is the AI engine underneath nearly every Microsoft AI product. Adding Claude as a native first-class option would mean offering Anthropic (an OpenAI competitor) inside Microsoft's own agent platform.

Where Claude does fit in the Microsoft ecosystem

Claude is available in Microsoft's broader cloud stack, just not in Copilot Studio's native picker:

Azure AI Foundry (formerly Azure AI Studio) added Claude models in late 2025 under the expanded Anthropic + Microsoft partnership. Available models include Claude Sonnet 4.6, Claude Haiku 4.5, and Claude Opus 4.8. Foundry is the engineering-team deployment platform — it's where you build custom Claude-powered applications with full Azure compliance, identity, and billing.

Azure AI Search + Claude grounding is supported for retrieval-augmented generation patterns where Claude generates and Azure handles the vector store.

Microsoft 365 Copilot (the consumer assistant inside Word, Excel, etc.) uses Microsoft's chosen models internally; users don't pick the underlying model.

Copilot Studio sits separately. It's the no-code agent builder layered on top of Power Platform — and its native AI engine is OpenAI-only. This is the gap.

The custom-connector workaround

To use Claude inside a Copilot Studio workflow, the supported pattern is a custom connector that wraps the Anthropic Messages API. Three steps:

1. Get a Claude API key. Two paths:

• Direct from Anthropic — sign up at console.anthropic.com, generate an API key, bill Anthropic directly. Cleanest for builders who want simple per-token pricing.
• Via Amazon Bedrock — if you're already on AWS, Claude is available through Bedrock with the same models at the same token rates plus your standard AWS billing integration. This works if your organization wants AWS compliance and billing for the Claude side, but you're using Microsoft Copilot Studio for the agent layer.

2. Build a custom connector in Copilot Studio. In the Power Platform admin center, create a new custom connector with:

• Base URL: https://api.anthropic.com/v1
• Authentication: API key in the x-api-key header, plus anthropic-version: 2023-06-01 and content-type: application/json
• Action: POST /messages with the standard Anthropic payload (model, max_tokens, messages array)

The connector definition is reusable across all your Copilot Studio topics. Build it once.

3. Call the connector from a Copilot Studio topic. Add an HTTP action node to your topic flow that invokes the custom connector. Parse the Claude response and route the assistant text into your conversation. The user sees a normal Copilot Studio conversation; the AI underneath is Claude.

The whole setup typically takes a developer 1-2 hours the first time. Subsequent topics that want Claude just reference the existing custom connector.

When Claude inside Copilot Studio actually makes sense

The custom-connector path is real but it's overhead. Three scenarios where it's worth the effort:

1. You're locked into Microsoft 365 for ops but Anthropic for AI quality. Your team builds workflows in Teams, SharePoint, Dynamics — Copilot Studio is the natural builder. But you've evaluated models and prefer Claude's reasoning or coding quality. The connector bridges the two.

2. You're running Claude on Bedrock for AWS compliance. Your organization standardized on AWS for cloud workloads but uses Microsoft 365 for productivity. Bedrock-hosted Claude billed through your AWS account fits the compliance story; Copilot Studio surfaces the agent inside Microsoft 365.

3. You want to A/B test Claude vs GPT-5 in a Copilot Studio workflow. Build the custom connector once, route a percentage of traffic to Claude, route the rest to native GPT-5. Compare conversion and quality metrics directly in the same workflow.

When Claude in Copilot Studio is the wrong path

Four scenarios where you should NOT route Claude through Copilot Studio:

1. The workflow is genuinely code-shaped. If the agent's job is reading files, writing code, opening pull requests — that's Claude Code territory. Copilot Studio's UI is designed for conversational business workflows, not codebase work. See Claude Code vs Microsoft Copilot Studio for the full head-to-head.

2. You need the simplest possible architecture. If you can use Azure OpenAI inside Copilot Studio and the quality is acceptable, that's the path. Native GPT-5 in Copilot Studio is one stack to maintain; Claude via custom connector is two. Pick simpler when capability is comparable.

3. You're not committed to Copilot Studio specifically. If you're early in the decision and Claude is your preferred model, Azure AI Foundry is the more natural Microsoft-stack home for Claude work. Foundry handles deployment, identity, and billing for custom Claude applications natively. Use Foundry for the build, surface inside Microsoft 365 via your own UI rather than Copilot Studio.

4. Your team has no developer to maintain the custom connector. The connector is a Power Platform asset that needs occasional maintenance (API version updates, error handling, retry logic). If the team is non-technical and there's no engineer on call, the connector becomes a maintenance hole. Stick with native GPT-5 inside Copilot Studio.

The cost math

Two billing streams stack:

Copilot Studio platform fees (regardless of which model powers the conversation):

• Standard tier: $200/month per 25,000 messages
• Increases stepwise for higher message tiers
• This pays for the Copilot Studio runtime, identity integration, Microsoft 365 surface, and audit logging

Claude API token costs (when you route through a custom connector):

• Claude Sonnet 4.6: $3 input / $15 output per million tokens
• Claude Opus 4.8: $5 / $25 per million
• Claude Haiku 4.5: $1 / $5 per million
• Daily-verified rates on the AI API pricing reference

A typical Copilot Studio workflow doing 25,000 monthly conversations on Claude Sonnet 4.6, with ~500 tokens of input and ~300 tokens of output per conversation:

Line item	Calculation	Cost
Copilot Studio platform fee	Standard tier	$200
Claude Sonnet input	25,000 × 500 / 1M × $3	$37.50
Claude Sonnet output	25,000 × 300 / 1M × $15	$112.50
Total monthly		~$350

The same workflow on native Azure OpenAI GPT-5.4 ($2.50 / $15 per million) costs:

Line item	Cost
Copilot Studio platform fee	$200
Azure OpenAI tokens (similar volume)	~$143
Total monthly	~$343

The cost difference between Claude and Azure OpenAI inside Copilot Studio is minor at most volumes. The decision is on model quality fit and team preference, not unit economics. Compare directly using the cost calculator.

What we expect to change

The relationship between Anthropic and Microsoft has expanded steadily across 2025-2026 — first the Azure AI Foundry availability, then Microsoft Foundry agents using Claude as a supported option. The trend line points toward Claude eventually appearing in Copilot Studio's native picker, but Microsoft has not publicly committed to a timeline. As of mid-2026, the custom-connector path is the supported answer.

If you're making a multi-year tooling decision, build the connector now and assume Microsoft will eventually add Claude natively. Your existing connector-based topics will keep working either way; you'll just have a cleaner native path when it ships.

The decision rule

1. Is your job code-shaped? → Claude Code, not Copilot Studio. The two operate at different layers; the comparison is here.
2. Are you on Microsoft 365 with operational workflows, and is Claude your preferred model? → Copilot Studio with a Claude custom connector. Plan ~2 hours for the developer setup; the connector is reusable across topics.
3. Are you early in the decision and committed to Claude? → Build on Azure AI Foundry instead. Native Claude support, full Azure compliance, more flexibility for engineering-led teams.
4. Are you committed to Copilot Studio but flexible on model? → Native Azure OpenAI is the simpler path. GPT-5.4 ($2.50/$15) handles most production workflows at comparable quality to Sonnet.
5. Are you running both, A/B testing Claude vs GPT-5? → Custom connector for Claude, side-by-side with native GPT-5 in the same topic flow. Real-world quality comparison data is worth more than abstract benchmarks.

What to read next

The Claude Code vs Microsoft Copilot Studio comparison covers the head-to-head decision when the choice is between Anthropic's terminal coding agent and Microsoft's no-code business agent platform. The Claude Opus vs Sonnet 4.6 routing guide covers which Claude model to pick if you do go the connector route. The AI API pricing reference tracks daily-verified per-token rates for every Claude tier plus GPT-5 family for direct comparison.

If you're earlier in the model-vendor decision, the best AI coding agents in 2026 and the 2026 AI agent shortlist cover the broader landscape across all 27 platforms we track.