⚙️ Lab note (added September 2026): every agent in this series was built and run in the new Copilot Studio experience — what Microsoft now calls the GitHub Copilot harness (GA August 2026). The classic “standard harness” is a different build experience with different behaviour — my standard-harness rebuild and full harness comparison are scheduled on this blog for late October 2026.
While building the agent from the first post in this series, I opened the tool catalog expecting connectors and found something else: a tab called Model Context Protocol (MCP), and inside it, sitting between Dataverse and SharePoint like just another row, an entry named Work IQ (Preview). That one row confused me for a solid hour — because Microsoft documentation describes Work IQ as a knowledge source, and here it was dressed as a tool. If you have been wondering how MCP, Work IQ, and plain old website knowledge relate to each other, this post is the untangling.
MCP Is the Plug
The Model Context Protocol is not a Microsoft product. It is an open standard — a shared way for any AI agent to discover what actions an external system offers (things like read_query or fetch_blob) and to call them. Think of it as the USB-C of the agent world: it says nothing about what is on the other end, only how to connect. That is why the Copilot Studio catalog can list wildly different systems on one tab — they all speak the same plug standard.
Work IQ Is the Appliance
Work IQ, on the other hand, is very much a product: the Microsoft intelligence layer over your Microsoft 365 work data — mail, calendar, Teams, OneDrive, files, people. It is the same layer that powers Microsoft 365 Copilot itself. When Microsoft wanted custom agents to reach it, they needed a delivery mechanism, so they shipped it as a set of MCP servers. The precise sentence, and the one I would put in an exam: Work IQ is a product that includes MCP servers as its agent interface — it is not simply an MCP.
Two practical tells give the relationship away. The connection mechanics follow the protocol: Work IQ sits in the same catalog, uses the same connection dialog, and raises the same Allow/Deny consent cards as every other MCP server. But the license follows the product: Work IQ requires a Microsoft 365 Copilot seat, because of what it accesses — not how it connects. The plug is free; the appliance is licensed.
And Websites Are Neither
Here is where the third mechanism sneaks in. The websites we added as knowledge in the first post involve no connection at all — no sign-in, no tools, no consent cards. The agent simply runs a web search scoped to the domains you allow and grounds its answer on what comes back, with citations. Read-only, public, passive. Knowledge is what the agent knows; tools are what the agent can touch.
Why the Confusion Exists
Because the product moved faster than the paper. The documentation still describes a Microsoft IQ knowledge panel with Work IQ toggles for mail and calendar — a UI my tenant simply does not have. What my tenant has instead is the MCP catalog above, including per-workload Work IQ servers (Mail, Calendar, Teams, OneDrive, SharePoint, User) for agents that need narrower scopes. When the docs and the product disagree, the product is telling you where the roadmap actually went: everything is converging on MCP as the one doorway for agent capability.
The Cheat Sheet
- MCP — the plug. An open protocol; how any agent connects to any system. Free, universal, boring in the best way.
- Work IQ — the appliance. The M365 work-data intelligence product, delivered to agents through MCP servers. Licensed by what it reaches (M365 Copilot seat).
- Dataverse / SharePoint MCP servers — more appliances, same plug. Covered hands-on in the next two posts.
- Website knowledge — neither. Passive, scoped web search with citations; no connection, no license beyond Copilot Studio itself.
Key insight: when someone asks whether to use MCP or Work IQ, the question is malformed — it is like asking whether to use USB or a kettle. The real questions are: what should our agent be able to read (knowledge), what should it be able to do (tools), and which appliances do those actions live in.
What I Learned
Getting this taxonomy straight before wiring anything saved me from a category error that would have leaked into architecture diagrams, license requests, and security reviews alike. I speak from experience: my own first diagram drew Work IQ in the knowledge column, and I had to redraw it once the product showed me otherwise. MCP is how agents connect; Work IQ is one of the things worth connecting to.
Next in this series: the Dataverse MCP server — where the agent stops reading and starts doing. (Licensing for all of this is in the credits post.)
