There’s a specific kind of friction that shows up in mature AI-adoption verticals. Professionals in these markets have already adopted AI tools — they’re not skeptics. But the tools they use live in a different context than the work they’re actually doing. The lease abstraction happens in one application. The financial analysis happens in Claude. The memo gets written in another tool. The outputs of each step don’t flow automatically into the next.

This is the interface gap: the distance between where AI-assisted tasks happen and where the professional’s actual workflow lives. It’s not a gap that better extraction accuracy closes, or more fields, or cheaper per-unit pricing. It’s a gap in where the tool lives relative to where the work lives.

The standalone SaaS model creates this gap structurally. A purpose-built lease abstraction tool extracts leases very well. But it exists as a separate application with its own login, its own interface, its own export flow. When a professional wants to use those results in Claude — to ask questions, run calculations, compare across properties, draft the investment memo — they have to close one window, open another, copy data across, and restart their context. The AI that did the extraction doesn’t know anything about the AI that’s doing the analysis.

MCP-native tools close this gap by living in the same environment as the AI assistant the professional already uses. The extraction and the analysis happen in the same session. The tool’s output is immediately available to the model for follow-on questions. There’s no export step, no context switch, no copy-paste. The workflow is continuous.

This matters most in workflows where the extraction is a prerequisite for the real work, not the end product. For a document-heavy professional workflow — lease review, deal analysis, due diligence — the goal isn’t a structured JSON of extracted fields. The goal is to underwrite the deal. Extraction is the mandatory first step. Everything else is downstream. If the tool that does the extraction is disconnected from the AI doing the underwriting, you’ve added friction at exactly the wrong moment.

The standalone SaaS incumbents can’t easily close this gap from the inside. They can ship an API, but that requires their customers to do integration work. They can build a Claude integration, but that’s a feature on top of a product that wasn’t designed for this context. The architectural starting point is wrong. An MCP-native tool starts from the context — Claude Desktop, the professional’s AI workspace — and builds the extraction capability there. The interface is the design constraint, not an afterthought.

For a professional who already uses Claude to do analysis work, an MCP tool for document extraction isn’t a new workflow. It’s the missing first step in a workflow they’re already running.

+++