Access customer service data and tools from MCP-compatible AI experiences — while minimizing the need for one-off custom integrations.

Today we’re announcing the general availability of the Dynamics 365 Customer Experience MCP Server for Service. Building on the public preview we released last winter, GA makes the service MCP tools and MCP apps used by Service Agent in Microsoft 365 Copilot available as an independent offering that organizations can connect to a growing ecosystem of MCP-compatible AI clients.

For business decision makers, the opportunity is clear: Agentic Customer Experience brings intelligence directly into the flow of work — helping teams access trusted data, surface relevant context, and support decision-making within existing workflows. With the Dynamics 365 Customer Experience MCP Server for Service, organizations can make Customer Service capabilities available in supported MCP-enabled experiences, so cases, knowledge, and customer context can be accessed within the tools teams already use.

1. What it is, and why “connect any client” matters

The Service MCP Server implements the Model Context Protocol (MCP), an emerging open standard for connecting AI applications to enterprise tools and data. Rather than developing separate integrations for individual AI experiences, organizations can use MCP to make Customer Service capabilities available to compatible clients through a standardized approach.

Manage connections through the Microsoft-hosted Agent 365 Tooling Gateway, which handles authentication to your Dataverse environment. Access aligns with existing Dataverse roles and permissions, helping organizations maintain established governance and access controls while supporting requirements for visibility, security, and governance.

Because it speaks a standard protocol, the server works with a broad and growing set of hosts, including:

• Microsoft 365 Copilot, including Service Agent, generally available on June 30th
• Microsoft Copilot Studio agents
• Visual Studio Code and GitHub Copilot CLI
• Other MCP-compatible clients that support HTTP-based MCP connections, including examples such as ChatGPT and Claude Code

The result is greater flexibility in how Customer Service capabilities can be made available across different AI experiences, while reducing the need to rebuild separate integrations for each interface.

Experience in ChatGPT:

Experience in GitHub Copilot CLI:

2. What’s available at GA

At general availability, the server ships with more than 90 service-oriented tools spanning many day-to-day workflows in a customer service organization. The capabilities cover common areas of service work, including:

• Case management — list, open, summarize, update, reassign, and resolve cases; pick the next case from a queue; and track SLA status and suggested next actions.
• Customer context — pull account and contact details and summaries, and review the activity timeline behind a case.
• Knowledge — search knowledge articles, draft and harvest new ones, check for knowledge gaps, and publish approved content.
• Email and activities — draft, generate, and send email; apply and manage templates; log notes and activities; link email to cases; and create a case directly from an email.
• Intelligence and recommendations — surface next-best-action suggestions, preview responses, and answer knowledge and service questions in context.
• Data and search — query and update Dataverse entity records and run cross-entity search and lookups.
• Environment and administration — select and manage MCP tool extensibility, connect to Copilot Studio agents, and customize Dynamics grids and views.

For your teams, that breadth means the server can support real, end-to-end workflows. From answering questions to taking the actions necessary to get work done.  

3. Extensible by design

The capabilities above are the starting point. The Customer Service MCP Server is built to be extended, with admin and maker tools that help organizations shape the experience around their business requirements:

• Connect your own external MCP servers. Admins can register, test, and manage additional MCP servers, so tools from other line-of-business systems can appear alongside built-in Customer Service capabilities in a governed surface.
• Bring in Copilot Studio agents. Admins can register and manage Copilot Studio agents and bots, extending the experience with custom agents already built in Copilot Studio.
• Tailor it to your teams. Maker and admin tools let organizations configure agent profiles, capabilities, and user preferences so the experience can better reflect established processes.

This extensibility is what makes the server a durable foundation rather than a fixed feature set. As your AI strategy evolves and new tools and agents come online, you can bring them into the same governed experience — no rip-and-replace required.

Getting started

The Dynamics 365 Customer Experience MCP Server for Service is generally available now. Administrators can connect it to Copilot Studio and supported MCP clients through the Agent 365 Tooling Gateway. Step-by-step setup, supported clients, and authentication details are available in the Microsoft Learn documentation.

Help teams access Dynamics 365 Customer Service capabilities within MCP-compatible tools and experiences.

The post Bring Dynamics 365 Customer Service to MCP-compatible AI clients: Customer Service MCP Server is now generally available appeared first on Microsoft Dynamics 365 Blog.

Watch your AI work on real, live cases, predicting, recommending, and resolving, without it ever touching a customer, a record, or a workflow. This is how you prove automation before you turn it on.

The hardest question in enterprise AI

Every team exploring AI automation eventually hits the same wall. The technology is ready. The use cases are obvious. The productivity upside is enormous. And yet one question stops the rollout cold:

“How do we know the AI is making the right decisions?”

It’s a fair question, and the most important one. In live customer cases, a wrong call isn’t a small error; it can mean a misrouted escalation, premature closure, or a customer feeling unheard.

Meet Shadow Mode

Shadow Mode in Case Management Agent lets you evaluate AI on live production cases, without taking any action.

The AI agent runs quietly alongside your human workflows. On real, incoming cases it observes, predicts, recommends, and simulates exactly what it would do if it were live. You see every recommendation, and the reasoning behind it, in real time. But nothing leaves the system : no record is updated, no customer is contacted, no status changes.

In other words, you get to watch your future automation work—on today’s hardest cases, before you ever hand it the keys.

Don’t debate whether the AI is production-ready—run it against live cases in shadow mode and let real outcomes validate performance. It won’t impact production data; it only observes and evaluates behavior.

Shadow Mode Behavior: What Happens—and What Never Happens

Shadow Mode is built on a simple promise: full visibility and controlled execution.

While Shadow Mode is active, the agent will:

• Run against live cases
• Surface recommendations and predicted actions in real time
• Show the reasoning behind every decision
• Measure readiness before you enable autonomous execution

And while it does all of that, it will never:

• Update a case record
• Send a customer communication
• Change a case status
• Interrupt an existing workflow

That separation is the whole point. You evaluate AI under real-world conditions while keeping complete operational control.

Validate across the entire case lifecycle

Shadow Mode isn’t a narrow test. It spans every major capability of Case Management Agent, so you can build confidence end to end.

Case Enrichment — Is the AI predicting cases correctly?

See how accurately the agent predicts the attributes that route and prioritize work: category, priority, product, and your custom fields and metadata. Compare predicted values against actual outcomes, pinpoint where accuracy slips, and sharpen your field descriptions before enrichment goes live.

Case Resolution — Would the AI solve it the way you would?

Evaluate AI-generated resolutions on real cases without sending a thing. Inspect how the agent identifies intent, retrieves knowledge, recommends a resolution, drafts the customer response, and shows its supporting rationale and evidence.

Case Follow-Up and Closure — Does the AI know when to act, and when to wait?

Pressure-test the lifecycle decisions that are easiest to get wrong: follow-up timing, SLA-based reminders, closure eligibility, and whether a case is truly ready to wrap. Catch premature closures and missed follow-ups in the safety of Shadow Mode—not in your customers’ inboxes.

From experiment to production—with confidence

Most validation relies on historical data that never quite captures how messy real cases get. Shadow Mode flips that. It measures your AI against the genuine complexity and unpredictability of active customer interactions, which means the readiness signal you get is the one that actually matters.

The Result:

• Real-world validation on live production data
• Transparent, explainable decision-making you can defend to stakeholders
• Side-by-side comparison with human actions
• Continuous improvement through iterative testing
• Less deployment risk, and a faster path to production

You move from “we think it’s ready” to “we’ve watched it perform” – and that’s the difference between hoping automation works and knowing it does.

Example: Shadow Mode evaluation for AI-generated case resolution

Start validating today

Don’t wait until production to learn how your AI will perform. Turn on Shadow Mode, let the agent work through live cases, and let the evidence build your confidence.

Get started in two steps

• Go to Copilot Service admin center > Case settings > Case Management Agent
• Activate Shadow Mode for the capability you want to validate.

The fastest way to trust automation is to watch it earn that trust. Shadow Mode is where it starts.

For step-by-step instructions: Enable Shadow mode in Case Management Agent

The post Trust before you automate: introducing Shadow Mode in Case Management Agent appeared first on Microsoft Dynamics 365 Blog.