Every article about Anthropic’s Claude For Legal launch led with the same number: twelve. Twelve role-specific legal plugins, covering Commercial, Employment, Privacy, Product, Corporate, AI Governance. The headlines wrote themselves. Legal-tech press fixated on the plugins. The plugins are visible. They have names. They have specific use cases. They are easy to describe.
The plugins are not the story.
The story is the other number in the announcement, the one that ran in the bottom half of most articles: more than twenty MCP connectors that ship alongside the plugins. DocuSign. Ironclad. iManage. NetDocuments. LexisNexis. Thomson Reuters. Box. Everlaw. Westlaw. Microsoft 365 integration covering Word, Outlook, Excel, and PowerPoint as a single context-carrying agent.
The plugins are what the marketing team wanted you to see. The connectors are what the strategy team built.
I want to write about why the second number matters more, because almost no one is writing about this correctly, and because the implications for the legal-tech industry are larger than the implications of any specific plugin.
What a connector actually is
Let me explain the distinction in plain terms, because the technical language obscures the importance.
A plugin is a specialized capability inside Claude—a configured persona that knows how to do a specific kind of legal work. The “Employment” plugin knows how to draft an employee handbook. The “Commercial” plugin knows how to review a vendor agreement. These are useful features. They are also features that any AI vendor with sufficient resources can build, given a few months and a team of legal subject-matter experts. There is no deep technical moat in a plugin. There is only product effort.
A connector is something structurally different. A connector is a permission for Claude to read, in real time, from a system the lawyer already uses. When Claude has a Westlaw connector, it doesn’t just “know about case law” from training data—it can actually query Westlaw, retrieve the actual current case, and ground its output in verified primary source material. When Claude has an iManage connector, it can read the law firm’s actual document management system, find the relevant precedent agreements, and work with them. When Claude has a DocuSign connector, it can route the document for execution after drafting it.
A plugin says “Claude knows how to do X.” A connector says “Claude is plugged into the systems where X actually happens.”
This is the difference between a smart consultant and a smart consultant who is already inside your firm.
Why this is the actual moat
I want to be careful here, because saying “this is the moat” sounds like cheerleading, and I’m trying not to do that.
But consider what the specialized legal AI companies—Harvey, CoCounsel, Spellbook, the others—have historically argued was their competitive advantage. The argument went, in essence, like this: we’re trained on legal data, we integrate with Westlaw, we have native connections to document management systems, we know the workflow. Buy us instead of trying to use a general-purpose tool, because we have all the pieces a lawyer actually needs, in one place.
Every single one of those advantages was a vertical-specific integration. They were valuable because the general-purpose alternatives didn’t have those integrations.
Anthropic just released twenty of them in one announcement.
In the technical sense, none of this is impossible for specialized vendors to maintain. They still have their connectors. The connectors didn’t disappear. But the exclusivity of the connectors disappeared. The thing that made vertical legal AI worth paying a premium for—the integrations general AI lacked—is no longer differentiated.
What happens next is a strategic problem for specialized vendors that I do not think they have fully absorbed yet. Their entire pitch has been “we have the integrations you need.” Now Claude has those integrations too. The pitch has to change.
The Microsoft 365 connector deserves its own paragraph
Among the twenty-plus connectors, one deserves special attention, because its strategic implications are larger than any of the others combined: the Microsoft 365 integration.
Here is what that means in practice. A lawyer is writing a contract in Microsoft Word. With Claude For Legal active, Claude is already there—not as a separate application the lawyer has to switch to, not as a tab in a browser, not as a tool that requires a context switch. Claude is inside the document, with awareness of the document, ready to help. The lawyer does not have to leave their writing environment. The AI shows up at the point of use.
The same is true of Outlook. The lawyer is responding to a client email. Claude is in the email client, with context on prior correspondence, ready to draft a response. The lawyer is not opening a separate tool to ask “how should I respond to this?” The AI is integrated into the email itself.
This is what zero-friction adoption looks like. It is not a feature. It is a structural change in how AI gets into legal work. And it is something specialized vendors fundamentally cannot match, because they would have to either build their own competing email client (impossible) or beg Microsoft for access (a permission Microsoft has not historically given lightly).
When I described the launch in my earlier piece as “the friction was the actual problem,” this was what I had in mind. The Microsoft 365 connector is what eliminates the friction. The plugins are what get the headlines. The integration is what wins the market.
The grounding part, briefly
I want to give the connectors one more credit before moving on, which is that they don’t just provide access—they provide the architecture for grounded output.
The single biggest fear about AI in legal practice has been hallucinated citations. Lawyers have been sanctioned for filing AI-generated briefs containing case names that don’t exist. Bar associations have issued warnings. The fear is justified and the consequences are real.
What the connectors enable is a structural answer to this fear. When Claude queries Westlaw for a relevant case, the case exists—it is being retrieved from Westlaw’s actual database, not synthesized from training data. When Claude pulls a document from iManage, the document is real. The output can be made traceable to specific source material, with citations the lawyer can click through and verify.
This is not a complete solution to the hallucination problem. Models can still misrepresent what a retrieved document says, or cite a real case for a proposition the case doesn’t actually support. But the worst failure mode—citations to cases that don’t exist at all—is structurally mitigated by grounding. For lawyers evaluating whether to use AI for client work, this is the difference between “interesting experiment” and “production deployment.”
The connectors are what make grounding possible. The plugins are what come after.
What specialized vendors should be doing now
If I were running a specialized legal AI company right now, I would be having an uncomfortable conversation with my team.
The conversation would start with: what is our defensible value, now that the general platform has our integrations?
The honest answers are not numerous. There are essentially three:
One: depth of specialization within a database. If you have not just access to Westlaw but a fundamentally better way of using Westlaw—proprietary workflows, custom retrieval logic, indexing that the general platform doesn’t replicate—you might have a moat. You will need to be very specific about what that depth looks like, and whether it is actually depth versus just polish.
Two: deep workflow integration with specific user habits. If your customers have developed years of muscle memory around your interface, and switching costs are high, you have temporary moat through inertia. But inertia is not a long-term strategy. The general platform is showing up in the same Microsoft 365 environment your customers already use. Your inertia advantage has a clock on it.
Three: horizontal expansion outside law. If you can connect your legal-specific intelligence to adjacent professional services—investment banks, consulting firms, financial advisors, accountants—you might escape the law-vertical squeeze by becoming a multi-vertical tool. But this dilutes your specialization, which was supposed to be your advantage in the first place. It is a real strategic option but it is a hard one.
I don’t see a fourth defensible position. The vertical companies are not dead, but the conversation about why they exist has fundamentally shifted, and the burden of proof is now on them rather than on the general platform.
What this means for lawyers, today
If you are a lawyer deciding what to use, the framing I would suggest is this: stop comparing products on features. Start comparing them on friction.
Where does the AI need to be? In your Word document. In your Outlook inbox. In your document management system. Connected to the legal databases you already query.
Which products meet you there? As of today, Claude For Legal does, more comprehensively than any specialized vendor. That is the assessment, and it is not a close one.
This will not stay still. The specialized vendors will respond. Some will adapt. Some will be absorbed. Some will carve out the depth-based niches I described above. In a year, the landscape will look different. But the directional reality is clear: the future of legal AI is integrated, in-workflow, grounded, and built on top of general platforms—not separated from them.
The headlines reported twelve plugins. The story was twenty connectors. And the story, as is often the case, was buried in the part of the announcement nobody read carefully.
This is the second in an ongoing series of reactions to Anthropic’s Claude For Legal launch. The first piece, on how the launch reshaped my view of the vertical-vs-horizontal question, appeared earlier this week.
If you’re at a specialized legal AI company and want to share, anonymously, how your team is thinking about this—email [email protected]. I read everything.