If you've spent any time around the AI conversation in the last year, you may have heard the term MCP — Model Context Protocol. It sounds technical. It is technical. But the thing it does is actually quite simple to explain, and if you use AI tools outside of FactSet, it's worth understanding.
What is MCP
AI tools like Claude, ChatGPT, and Microsoft Copilot are powerful — but they're only as useful as the data they have access to. By default, a general-purpose AI tool doesn't know anything about your company's ownership structure, your peer group's recent earnings calls, or the analyst estimates that crossed your desk this morning.
MCP is a standardized protocol — a shared technical language — that allows external AI tools to securely connect to a trusted data source and pull from it in real time. FactSet has built an MCP server that does exactly this: it gives AI tools a direct, structured connection to FactSet's data.
When you ask Claude or ChatGPT a question that requires financial data, and FactSet's MCP server is connected, the AI tool doesn't guess or draw on stale training data. It queries FactSet directly and returns an answer sourced from the same data you'd find in the platform.
Why MCP is different from in-platform AI
The AI tools embedded in Irwin and FactSet — Transcript Assistant, Irwin AI Summaries, AI-powered document search — are purpose-built for specific tasks within the platform. They're excellent at what they do.
MCP is a different kind of capability. It's not about replacing the in-platform experience. It's about meeting you where you already work.
If you've started building workflows around AI tools you use daily — drafting in Claude, organizing in Copilot, researching in ChatGPT. Without MCP, those workflows run on whatever data the AI tool already has, which is often general, dated, or simply wrong on financial specifics. With MCP, those same workflows can draw on FactSet's data directly. You don't have to choose between the tool you already know and credible financial data.
What this looks like in practice
- Drafting investor communications. You're in Claude writing a shareholder letter and you want to reference how analyst consensus on your sector has shifted over the past two quarters. Without MCP, you'd tab out of your writing workflow, pull the data from FactSet, then come back and paste it in. With MCP, you ask Claude directly — and it pulls the FactSet data without you leaving the conversation.
- Preparing for a one-on-one investor meeting. You're using Copilot to pull together a briefing document before an investor meeting. You want to include recent ownership changes and what the investor's portfolio has been doing. With FactSet's MCP server connected, Copilot can surface that ownership and holdings context from FactSet as part of building the document.
- Peer benchmarking on the fly. You're in ChatGPT working through a peer comparison and want current estimates and ratings for four companies in your sector. Instead of pulling each one manually from the Workstation, MCP lets the AI query FactSet and return the comparison directly.
- Board prep. You're drafting a board presentation and want to show how analyst estimates for your company have moved relative to peers over the past year. FactSet's estimates data — the same data your investors use — comes in through MCP, sourced and citable, without a separate research step.
What FactSet data is accessible through MCP
Through FactSet's MCP server, AI tools can access data categories that power your daily IR workflow — ownership, consensus estimates, fundamentals, and pricing – alongside transcripts, Street Account, and filings. The same data that powers your Workstation is queryable from inside the AI tools you already use.
What it takes to set up MCP
MCP setup doesn't require a technical background or developer resources. For Claude and ChatGPT, setup happens directly in the tool. For Microsoft, the integration runs through Copilot Studio – once configured there, the MCP-enabled agent can be published to standard Microsoft 365 Copilot, which is where your team will actually use it day-to-day. Studio access is the prerequisite; once it’s in place, the agent rollout is straightforward.
Your Irwin or FactSet account team can walk you through setup for your specific configuration — it's typically a short process, not a project.
The question to ask your vendors
Not every AI tool in your stack is MCP-enabled, and not every data provider has built an MCP server. As you evaluate how AI fits into your IR workflow, the right question is: what data does this actually pull from, and can I verify it?
MCP gives that question a clear, structural answer. The source is FactSet. The connection is standardized. The output is traceable.
That's not a feature. That's a foundation.
