Microsoft Copilot Studio: What It Is and When to Use It

Microsoft Copilot Studio is a low-code web studio, at copilotstudio.microsoft.com, for building AI agents and automated workflows, connecting them to your organization’s data and systems, and publishing them where people already work: Teams, Microsoft 365 Copilot, a website, a mobile app and more. Use it when an agent has to serve a department or customers, call business systems through connectors or MCP servers, run multi-step processes, or be governed with separate test and production environments. If you only want a quick question-and-answer agent over your own files for yourself or a small team, Agent Builder inside Microsoft 365 Copilot is simpler and usually enough. What follows is what Microsoft documents as of September 30, 2026.

What Copilot Studio builds

Microsoft’s Copilot Studio overview↗ (opens in a new tab) names three building blocks, which you can use alone or combine:

• Agents: AI assistants that follow your instructions, draw on the knowledge sources you connect, and use tools to act. You can describe one in plain language, test it, then publish it. Some agents can be given their own account so they work proactively, for example on a new employee’s onboarding.
• Workflows: automations built in a drag-and-drop designer, where each step can reason and act, with built-in testing and human-in-the-loop controls.
• Agent flows: the established flow format, similar to Power Automate. They run on their own or attach to an agent as a tool, and can include human review steps.

Two newer pieces are marked preview on that page: building from a plain-language description of a business goal, and the app creation experience. Treat both as subject to change.

Harnesses: how the agent reasons

Everything in Copilot Studio runs on a harness, the runtime between your design and the model. According to Microsoft’s page on harnesses in Copilot Studio↗ (opens in a new tab), the choice affects how the agent reasons, what it can do out of the box, and how it is billed:

• GitHub Copilot harness: for reasoning-heavy, multi-step work. It breaks a goal into steps, calls tools across connectors, knowledge, MCP and other agents, creates and edits Word, Excel, PowerPoint and PDF files, and supports skills and memory.
• Standard harness: for rule-based agents and structured conversations built from topics, the designed conversation paths Copilot Studio has long used.
• Copilot chat harness: for extending Microsoft 365 Copilot Chat with your organization’s knowledge, published to internal teams only.

The name causes confusion. Microsoft states that the GitHub Copilot harness is a Copilot Studio framework, not the GitHub Copilot service, and that customer data is not sent to GitHub Copilot when agents run. Copilot Studio is also not GitHub Copilot’s agent mode in VS Code or Visual Studio, which is a coding assistant; that is covered in GitHub Copilot agent mode best practices.

Knowledge, tools and connectors

An agent answers from knowledge and acts through tools. Most tools come from Power Platform connectors, the same wrappers around APIs that Power Automate and Power Apps use:

• Standard connectors, such as SharePoint, come with every Copilot Studio plan. Premium connectors are available in select plans.
• Custom connectors reach any public API that no prebuilt connector covers.
• You add a connector tool at the agent level under Tools, or inside a topic. By default it asks each user for their own credentials; you can switch a tool to maker-provided credentials, which means anyone who can chat with the agent borrows your access.

MCP in Copilot Studio

Copilot Studio agents can also use tools from a Model Context Protocol server. Microsoft’s page on extending an agent with MCP↗ (opens in a new tab) says it supports MCP tools and resources, not prompts, and that generative orchestration must be on. Tools the server adds, changes or removes show up in the agent automatically.

• Transport: Streamable HTTP only. Microsoft says SSE is no longer supported after August 2025, so an older server that only speaks SSE will not connect.
• Sign-in: none, an API key sent as a header or query parameter, or OAuth 2.0. OAuth has three modes: dynamic discovery for servers that support dynamic client registration with discovery, dynamic with endpoints you enter, and manual with a client ID and secret.
• Governance: MCP servers ride on Power Platform connectors, so a data policy that governs connectors governs MCP servers too.
• Catalog servers: some vendors appear in the list when you add an MCP tool. Box’s guide to setting up the Box MCP server in Copilot Studio↗ (opens in a new tab) has a Box admin make the Copilot Studio integration available first, then the maker picks Box from the Model Context Protocol list and signs in.

The step-by-step for adding any remote server, with a worked example, is in Microsoft 365 Copilot agents: examples and how to build one. If you are choosing between a local and a hosted server, Copilot Studio needs a hosted one; local vs remote MCP servers explains the difference.

Where Copilot Studio agents run

You publish once and connect channels: Teams and Microsoft 365 Copilot, SharePoint, WhatsApp, a demo website for testers, your own website, a mobile app, Facebook, and further channels through Azure Bot Service. Agents authenticate with Microsoft Entra ID by default. Choosing No authentication lets anyone with the link chat with the agent, and such an agent cannot use tools that need user credentials. Admins can turn channels off for the whole organization.

Licensing, by name

Microsoft’s Copilot Studio licensing page↗ (opens in a new tab) describes usage in Copilot Credits, which replaced messages as the unit on September 1, 2025. The ways in:

• A Copilot Studio user license, which is free of charge but needs a tenant prepaid Copilot Credit pack subscription first.
• Pay-as-you-go through an Azure subscription, or a one-year Copilot Credits prepurchase plan.
• A Microsoft 365 Copilot license, which lets you use Copilot Studio to extend Microsoft 365 Copilot with agents.
• A trial license, which lets you build and test but not publish.

Microsoft’s pages are not quite aligned here. Its comparison page says both Agent Builder and Copilot Studio are included with a Microsoft 365 Copilot license. The licensing page is narrower: the license covers agent capabilities used in Microsoft 365 Copilot, Teams or SharePoint, and zero-rates certain kinds of answers there. An agent on your website, or one doing other kinds of work, uses Copilot Credits. Check the current Copilot Studio Licensing Guide with whoever buys your licenses before you plan a rollout.

Copilot Studio vs Agent Builder

Microsoft’s guide to choosing between Agent Builder and Copilot Studio↗ (opens in a new tab) draws the line by audience and depth. Agent Builder (which older guides call Copilot Studio agent builder or the lite experience of Copilot Studio) lives inside the Microsoft 365 Copilot app and suits information workers building Q&A agents over SharePoint files, email and other Microsoft 365 content for themselves or a small team. Copilot Studio suits makers and developers building for a department, the whole organization or customers, with multi-step logic, approvals, prebuilt and custom connectors, autonomous capabilities, and development, test and production environments. Agent Builder agents are managed in the Microsoft 365 admin center; Copilot Studio agents in the Power Platform admin center.

You do not have to choose forever. An agent made in Agent Builder can be copied to Copilot Studio, keeping its configuration and instructions, when it outgrows the simple tool.

When to use Copilot Studio, and when not to

• Use it when the agent must act in another system, such as creating a ticket, updating a CRM record or starting an approval.
• Use it when the audience is wider than your team, or includes customers.
• Use it when you need versioning, separate environments, analytics and evaluations before an agent goes live.
• Skip it for a personal helper over your own documents. Agent Builder does that in minutes.
• Skip it if your organization does not run on Microsoft 365. Most of its value comes from Entra sign-in, Teams, Microsoft Graph and Power Platform governance.
• Skip it for coding work. Coding agents such as GitHub Copilot and Claude Code belong in the editor and terminal.

Connecting a Copilot Studio agent to a task board

A common first job for an agent in Teams is “file this as a task.” fenbs is a task board where people and AI assistants are members with roles, and its remote MCP server at https://fenbs.ai/api/mcp publishes the OAuth discovery and dynamic client registration endpoints that the Dynamic discovery option looks for. Each person who uses the agent signs in to fenbs and ticks what it may do: reading the board is always on, and adding or changing tasks and commenting are theirs to choose. The access token lasts an hour and refreshes, and the connection can be revoked in Settings at any time.

The agent should call fenbs_whoami and then fenbs_get_context, which returns the rules from the Decisions and rules page first and the AI context notes after. It files work with fenbs_create_item as a feature, enhancement or bug with a note, a plan and a priority from 1 to 10, where 1 is the most urgent, and moves tasks through To Do, Next Up, In Progress and Completed. Every change is recorded in History with who made it. Be clear about the gaps: fenbs has no due dates, no settable assignee and no sprints, so an agent that is asked for a deadline or an owner should write it in the task note.