MCP is everywhere right now. But there’s a version of this story that stops too soon, and most of the industry is telling it.

There’s a lot of noise about which platform supports MCP, which AI tools connect to which systems, and who announced what this week. That’s all worth knowing. It’s just not the part that determines how useful legal AI actually is for your organization.

What MCP is, and why it matters

MCP, or Model Context Protocol, is an open standard from Anthropic that lets AI tools connect to outside systems through a single governed interface. People call it the USB-C for AI: one universal connection that replaces a tangle of custom integrations. In legal, it’s how an AI assistant reaches the documents, matters, and context that live inside the systems firms rely on most.

Before MCP, every integration was a one-off build. Each vendor had its own API, its own authentication, its own quirks. MCP changes that. It’s a shared language. And it moved fast. Anthropic introduced the spec in late 2024, and within months OpenAI, Google, and Microsoft had all adopted it. When the biggest players in the industry converge on a single protocol that quickly, it stops being one vendor’s idea and becomes infrastructure.

That’s the good news. But “MCP-compatible” is a starting line, not a finish line. And here’s where the story gets more interesting.

Connected to what, exactly?

Here’s what the protocol doesn’t tell you: MCP specifies how AI tools connect. It doesn’t specify what they get when they do. That part is entirely up to the system on the other end of the connection, and it’s the part that actually determines how useful the AI is.

Once every platform supports MCP and every AI tool can connect, the question shifts. The AI is connected. But connected to what? What does it actually receive when it reaches into your system? What does it know when it starts working?

There’s a meaningful difference between an AI that can retrieve a file and an AI that understands what that file means. Retrieval finds documents. Search finds concepts. Neither one tells an AI agent how a matter actually fits together: who the parties are, what came before, which clauses have precedent, what risk looks like in this context versus another one. That’s not a retrieval problem. That’s a context problem.

And that’s the question worth asking of every MCP integration: what does the AI really get?

There’s a second question hiding behind the first: where do your documents end up? Some MCP integrations work by copying or syncing your documents into the AI tool, which leaves you with a second copy to secure and govern, and the governance drift that follows. Others, including ours, leave the documents in place and deliver only the context the AI needs. Same protocol, very different risk posture. Not all MCP is equal.

What NetDocuments surfaces that others can’t

Most systems store your work. NetDocuments is built to understand it.

The legal context graph powers the NetDocuments experience in private preview today, and our MCP, also in private preview, carries that firm context to the AI tools you connect. It’s not a document index. It’s a map of relationships across your entire firm’s knowledge: documents connected to matters, matters connected to parties, parties connected to history, history connected to the work your firm has done for decades. An AI agent connected this way isn’t drawing from a simple file store. It draws on a governed, permission-aware understanding of what those files mean together, and that understanding deepens as we expand the surface. We’ve tested this, and that distinction produces more accurate results at lower cost than approaches that hand the AI a stack of documents and ask it to figure things out.

That depth already shows up today inside NetDocuments. That’s what makes Smart Answers work the way it does: answers grounded in your firm’s actual matter history, complete with citations, not generic outputs from a model that has never seen your precedents. More data isn’t the same as more understanding. The right context is. And it’s what extends through MCP to Claude, ChatGPT, and the other tools your firm already uses, so the context travels with the connection.

None of this is hypothetical. Smart Answers is live today, and our MCP is available now, carrying context into Claude, ChatGPT, and external tools your team works in. And the legal context graph as a fully realized layer across every connection is what we’re surfacing – deliberately, with the foundation already in place.

What to actually ask when evaluating this

If you’re looking at MCP integrations right now, three questions are worth putting to every vendor, including us:

• Does the AI get real context, or just files? Does the integration understand relationships across a matter, or does it retrieve documents and call it done?
• Does governance travel automatically? Are your ethical walls, permissions, and audit controls enforced on every AI interaction, or does someone have to rebuild that in a separate system?
• Can you audit every AI interaction that touched client content? Every AI request that touches client matter content is potentially subject to compliance review, ethical scrutiny, or discovery. If you can’t trace what the AI accessed, when, and on whose behalf, that’s not governance. That’s faith.

NetDocuments already answers those questions, with more coming in the months ahead.

And the same context that powers all of that inside NetDocuments extends to AI tools you connect through ndConnect, whether your team works in Claude, ChatGPT, or more. Your documents stay in NetDocuments, fully governed.

---

Learn more about how this works in practice, including which tools NetDocuments supports and how context travels through the connection, visit our Legal AI Platform page.

---