From Connectors to Context: What MCP Really Means for the Agentic Workplace (Part A)
Jan 08, 2026
Why the Agentic Workplace Needed a New Standard
For years, enterprise AI on Microsoft’s stack has relied on connectors, custom connectors, and flows to make systems talk to each other. These approaches work well for predictable automation, but they start to break down when you introduce agents—systems that need to reason, discover tools, and act dynamically across multiple environments. That’s where the Model Context Protocol (MCP) comes in.
MCP, introduced by Anthropic in late 2024, is an open standard that changes how AI agents connect to external tools and data. Instead of hard‑coding integrations, MCP gives agents a way to discover what tools a system offers, understand the inputs and outputs, and invoke them reliably at runtime. Microsoft adopted MCP in Copilot Studio in March 2025 and made it generally available at Build 2025, adding enterprise‑grade features like tool listing, streaming support, and detailed runtime tracing. MCP servers appear inside Copilot Studio through the Power Platform connector infrastructure, which means all the governance you already rely on—virtual networks, data loss prevention, and authentication—remains intact.
The Integration Problem: M×N vs M+N Explained
Before MCP, connecting agents to systems was a classic many‑to‑many problem. Imagine you have five different agents and ten different systems. If each agent needs to integrate directly with each system, you end up with fifty separate connections. That’s what we mean by M×N—M agents multiplied by N systems. It’s complex, expensive, and hard to maintain.
MCP changes this equation to M+N. Each agent only needs to know how to speak MCP, and each system only needs to expose an MCP server. Instead of wiring every agent to every system, you wire each agent to MCP and each system to MCP. This dramatically reduces complexity and makes integrations far easier to scale. In plain English: rather than building fifty separate bridges, you build one standard highway that everyone can use.
MCP’s Origin Story and Why It Matters
MCP is a client–server protocol that defines how agents discover and use three core elements: tools, resources, and prompts. Tools are functions the agent can call, resources are file‑like contexts the agent can read, and prompts are templates optimised for specific tasks. The protocol specifies how these are described, discovered, and invoked, using standardised schemas and transport options like JSON‑RPC. The goal is interoperability—agents and systems can connect without bespoke plumbing.
For agent makers, this matters because it eliminates brittle action lists. Instead of hard‑coding what an agent can do, you publish those capabilities once on an MCP server. The agent then discovers them dynamically and stays up to date as the server evolves. This is a huge shift from the connector era, where every change meant manual updates.
Microsoft’s Implementation in Copilot Studio
Microsoft introduced MCP support in Copilot Studio during the March 2025 preview and made it generally available at Build 2025. The implementation is pragmatic: MCP servers are surfaced through the existing connector infrastructure, so you keep enterprise security and governance while enabling agent‑native discovery. At GA, Microsoft added features like tool listing, streaming transport, and enhanced tracing so makers can see exactly which MCP tool executed at runtime.
Inside Copilot Studio, MCP sits within Generative Orchestration. When enabled, an agent can discover the tools and resources published by an MCP server and decide which ones to call based on intent and context. Inputs and outputs are inherited automatically, and updates on the server flow through without manual re‑binding. This is fundamentally different from traditional connectors, which expose fixed actions that makers must wire in advance.
Out‑of‑the‑Box MCP Experiences
Microsoft has published official labs and Learn documentation to help makers get started with MCP. One of the most significant out‑of‑the‑box experiences is the Dataverse MCP Server. Instead of wiring multiple Dataverse connector actions or building custom flows, you can add Dataverse as an MCP server and let the agent list tables, describe schemas, read and update records, and execute prompts—all governed inside the Microsoft estate. This reduces integration friction and accelerates build time.
The marketplace for MCP‑enabled servers is also growing, giving makers access to agent‑ready tools without reinventing integration code. This signals a clear direction: MCP is becoming a first‑class citizen in the Power Platform ecosystem.
Is MCP Just Another Connector?
It’s tempting to think of MCP as a new connector, but that misses the point. Connectors expose fixed actions that you add manually to topics or flows. MCP exposes a dynamic catalogue of tools and resources that agents discover and reason over at runtime. MCP servers can update their toolset without republishing an agent, which is not how traditional connectors work. By placing MCP inside the Power Platform, Microsoft avoided creating a separate governance story. You enforce data loss prevention, virtual network isolation, authentication, and ALM the same way you do for connectors, while enabling agentic behaviours that were previously impossible.
MCP vs RAG: Doing vs Knowing
It’s easy to confuse MCP with Retrieval‑Augmented Generation (RAG) because both involve giving AI systems access to external information, but they solve different problems.
RAG is about knowledge injection. It improves a model’s answers by retrieving relevant documents or chunks from a knowledge base and feeding them into the prompt. RAG doesn’t give the model new actions—it just gives it more textual context to reason over. It’s great for answering questions, summarising content, and compliance checks, but it doesn’t execute business logic or update systems.
MCP is about capability. It gives the agent a standard way to discover and invoke tools and resources so it can perform real operations—update a record, trigger a workflow, fetch structured data. In plain English: RAG helps the agent think better; MCP helps the agent do more.
The two approaches complement each other. An agent could use RAG to pull policy documents or case notes for reasoning, then use MCP to execute the next step—such as creating a compliance report or updating a case in Dataverse. RAG enriches the agent’s thinking, MCP expands its doing.
Where MCP Fits Functionally in Copilot
MCP is the mechanism that allows Copilot agents to discover and use capabilities from external systems in a generative way. Rather than hard‑coding a set of actions, the agent selects MCP tools based on intent and context. As your MCP server evolves, the agent’s capability surface evolves with it—without re‑authoring actions. This makes MCP central to Microsoft’s vision for multi‑agent orchestration, where agents delegate tasks to other agents across Microsoft 365, Azure AI, and Fabric. MCP provides the standardised capability layer that makes this orchestration practical.
The Bottom Line
MCP is not just another connector. It’s a standardised capability plane that makes agentic behaviours practical and maintainable in the enterprise. It delivers discovery, typed tools, live updates, and governed access—all inside the Power Platform’s security envelope. For makers, the value is clear: less glue code, more capability, and a cleaner path to scaling agents across the workplace.
In Part B, we’ll explore what happened when we applied MCP in Injury Guard AI and Safe Havens AI—where it accelerated build time, where it reduced maintenance, and where we still rely on Flows and AI Builder for compliance and document‑heavy processes.
WorkDynamics is here to help. Our eBook, Designing the Future of Work: Agentic Solutions for Scalable Digital Labour, offers twelve practical approaches to activating digital labour, modernising legacy systems, and building agentic workplaces. It includes real-world examples, frameworks, and access to ready-to-deploy solutions.
