How to access Copilot Studio agent definitions and metrics via API or SDK?

Question

How to access Copilot Studio agent definitions and metrics via API or SDK?

Nachiketha N 0

I'm working with Microsoft Copilot Studio and want to programmatically access agent definitions (e.g., name,tools, triggers,knowledgebase ,instruction) and usage metrics.

Are there supported Dataverse tables, Web APIs, or SDKs (e.g..NET SDK) that expose this?

If not yet available, is this on the roadmap or in preview

Karan Shewale 170 Reputation points Microsoft External Staff

2025-06-27T05:38:36.1766667+00:00

Thanks for bringing this issue to our attention. We will check and update you soon.

1 answer

Your answer

Karan Shewale 170 Reputation points Microsoft External Staff

2025-06-27T05:38:36.1766667+00:00

Thanks for bringing this issue to our attention. We will check and update you soon.

Answer 1

Karan Shewale 170 Microsoft External Staff

Hi Nachiketha N,

Thanks for reaching out about programmatic access to Copilot Studio data. Here's what's currently available and some guidance on next steps:

Current Options Available:

Dataverse Tables - Your best bet right now

Copilot Studio agents are stored in Dataverse tables (bot, botcomponent, botcomponentcollection)
You can access agent definitions including name, instructions, knowledge bases, and triggers
Use standard Dataverse Web API with Azure AD authentication

Power Platform Web API

RESTful endpoints for querying bot-related tables
Supports most CRUD operations for agent definitions
Good documentation available in Microsoft docs
Power Platform CLI & PowerShell
pac copilot commands for basic operations
Useful for automation scripts and bulk operations

Current Limitations:

Usage metrics and analytics have limited API access (this is the main gap)
Conversation logs are restricted for privacy/compliance reasons
No dedicated .NET SDK for Copilot Studio yet (though standard Dataverse SDK works for table access)

Recommended Approach: Start with the Dataverse Web API to access your agent definitions. For metrics, you'll need to work with what's available in the Power Platform admin center APIs for now.

Official Microsoft Documentation:

[Dataverse Web API Reference] https://learn.microsoft.com/en-us/power-apps/developer/data-platform/webapi/overview
[Power Platform CLI Reference] https://learn.microsoft.com/en-us/power-platform/developer/cli/introduction?tabs=windows
[Dataverse .NET SDK] https://learn.microsoft.com/en-us/power-apps/developer/data-platform/org-service/overview

Roadmap: Microsoft is actively expanding these APIs, particularly around analytics and monitoring capabilities. I'd recommend checking their quarterly platform updates for new releases.

Let me know if you need help with specific implementation details.

Thanks,

Karan Shewale

*************************************************************************

If the response is helpful, please click "Accept Answer" and upvote it. You can share your feedback via [Microsoft Teams Developer Feedback] link. Click here to escalate.

Nachiketha N 0 Reputation points

2025-06-30T06:44:32.41+00:00
I'm looking to understand the capabilities of Copilot Agent Studio in several areas

Does Copilot Agent Studio support versioning of agents and rollback configurations?

How are guardrails defined, implemented, and enforced within agent?

Is there support for structured input and output for agents, such as schema definitions for inputs and outputs?

How is an agent invoked, and how are input/output schemas associated with the invocation process?

What mechanisms are available for authentication and authorization of tools integrated with the agent?

What is the lifecycle of an agent and its related entities?

What are the different states an agent (or related entity) can be in, and what do those states represent?
Karan Shewale 170 Reputation points Microsoft External Staff

2025-07-01T05:39:46.1+00:00
Hi Nachiket, please find the answers to your questions below.

Does Copilot Agent Studio support versioning of agents and rollback configurations?

Agents are packaged and managed via Power Platform Solutions (managed/unmanaged), which support version tracking using Semantic Versioning (MAJOR.MINOR.PATCH) ([rishandigital.com](https://rishandigital.com/ai/versioning-in-copilot-studio-applications/?utm_source=chatgpt.com"Versioning in Copilot Studio applications - Rishan Solutions")).

You can export/import solutions manually or programmatically (via CLI) to manage versions and restore previous states—effectively rolling back by re-importing older solution packages ([rishandigital.com](https://rishandigital.com/ai/versioning-in-copilot-studio-applications/?utm_source=chatgpt.com"Versioning in Copilot Studio applications - Rishan Solutions")).

How are guardrails defined, implemented, and enforced within agents?

Governance is controlled via Microsoft 365 and Power Platform tools: Data Loss Prevention (DLP) policies, environment roles, sensitivity labels, and Purview compliance—for both design-time and runtime enforcement .

Microsoft also includes built-in guardrail mechanisms such as content filters, misuse detection, and policy monitoring to prevent inappropriate actions ([github.com](https://github.com/MicrosoftDocs/microsoft-365-docs/blob/public/copilot/copilot-tuning-troubleshooting-guardrails.md?utm_source=chatgpt.com"microsoft-365-docs/copilot/copilot-tuning-troubleshooting-guardrails.md ...")).

Is there support for structured input and output for agents, such as schema definitions for inputs and outputs?

Yes. Using the Power Platform authoring canvas, agents support typed input/output parameters (string, number, boolean, date), allowing structured I/O per topic or flow ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-managing-topic-inputs-outputs?utm_source=chatgpt.com"Manage topic inputs and outputs - Microsoft Copilot Studio")).

Advanced settings like entity recognition, validation rules, reprompt behavior, and default values are available for finely controlled input handling ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-additional-settings-topic-action-inputs?utm_source=chatgpt.com"Additional settings for inputs of topics and actions - Microsoft ...")).

How is an agent invoked, and how are input/output schemas associated with the invocation process?

Invocations include:

Chat-based triggers (via Copilot in Teams or M365),

**Power Automate flows** or scheduled/API triggers. Input variables are mapped at invocation; output variables are captured and passed along to downstream nodes or systems ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-flow-input-output?utm_source=chatgpt.com"Use input and output variables to pass information - Microsoft Copilot ...")).

What mechanisms are available for authentication and authorization of tools integrated with the agent?

Authentication is via Microsoft Entra ID (Azure AD), using user or service principal credentials.

Access control is governed through DLP policies, environment security roles, and connector permissions.

Agents enforce scoped access depending on who is invoking or publishing them (e.g., Org‑wide, group, or user-specific access) ([adoption.microsoft.com](https://adoption.microsoft.com/files/copilot-studio/Agent-governance-whitepaper.pdf?utm_source=chatgpt.com"Agent Governance Whitepaper - adoption.microsoft.com"), [community.powerplatform.com](https://community.powerplatform.com/forums/thread/details/?threadid=deb9d736-d79a-ef11-8a6a-7c1e521a8275&utm_source=chatgpt.com"Version History and Roll Back - community.powerplatform.com")).

What is the lifecycle of an agent and its related entities?

The lifecycle spans from authoring to decommissioning:

Authoring & Drafting – Design agent via studio UI.

Publishing/Creating – Finalize and activate agent ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/copilot-studio-agent-builder-build?utm_source=chatgpt.com"Use the Copilot Studio Agent Builder to Build Agents | Microsoft Learn"), [learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/copilot-studio-agent-builder-publish?utm_source=chatgpt.com"Publish and Manage Copilot Studio Agent Builder Agents")).

Sharing – Grant access to users/groups ([microsoft.com](https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/empowering-makers-with-a-complete-agent-lifecycle-in-microsoft-copilot-studio/?utm_source=chatgpt.com"Empowering makers with a complete agent lifecycle in Microsoft Copilot ...")).

Evaluating & Monitoring – Use built-in evaluation and analytics dashboards ([microsoft.com](https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/empowering-makers-with-a-complete-agent-lifecycle-in-microsoft-copilot-studio/?utm_source=chatgpt.com"Empowering makers with a complete agent lifecycle in Microsoft Copilot ...")).

Updating & Versioning – Export/import or use CLI/Git to update and iterate .

Rolling Back – Re-import prior solution version if needed ([rishandigital.com](https://rishandigital.com/ai/rollback-strategies-in-copilot-studio-deployments/?utm_source=chatgpt.com"Rollback strategies in Copilot Studio deployments")).

Decommissioning – Unpublish or delete agent at end of lifecycle.

What are the different states an agent (or related entity) can be in, and what do those states represent?

By Solution Type:

Unmanaged – Editable in development.

**Managed** – Locked for production environments ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/copilot-studio-agent-builder-publish?utm_source=chatgpt.com"Publish and Manage Copilot Studio Agent Builder Agents"), [adoption.microsoft.com](https://adoption.microsoft.com/files/copilot-studio/Agent-governance-whitepaper.pdf?utm_source=chatgpt.com"Agent Governance Whitepaper - adoption.microsoft.com"), [rishandigital.com](https://rishandigital.com/ai/versioning-in-copilot-studio-applications/?utm_source=chatgpt.com"Versioning in Copilot Studio applications - Rishan Solutions"), [learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/?utm_source=chatgpt.com"Microsoft Copilot Studio guidance documentation")). **Authoring/Execution States**: **Draft** – Under development. **Published/Live** – Accessible by users ([learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-365-copilot/extensibility/copilot-studio-agent-builder-publish?utm_source=chatgpt.com"Publish and Manage Copilot Studio Agent Builder Agents")). **Updated** – New version created. **Rolled back** – Reverted to a prior version. **Deleted/Deprecated** – Removed or no longer maintained. **Evaluated** – Tested during pre-launch ([rishandigital.com](https://rishandigital.com/ai/rollback-strategies-in-copilot-studio-deployments/?utm_source=chatgpt.com"Rollback strategies in Copilot Studio deployments"), [github.com](https://github.com/copilot-coding-agent/user-feedback/issues/38?utm_source=chatgpt.com"Agent keeps rolling SDK version backward in global.json as ... - GitHub")).

Share via

How to access Copilot Studio agent definitions and metrics via API or SDK?

1 answer

Your answer