Utah Law Firm Goes AI Native
Last week a US law firm started selling commercial contract redlines for $117, with attorney sign-off included and a 24h turnaround. The firm, one of the first inside the Utah Supreme Court’s sandbox built from the ground up around AI, partnered out of the gate with the Associated General Contractors of America. A managing partner in Chicago, Dallas or Miami who counts a construction GC among their clients is, as of this week, competing with a number that GC didn’t have available a fortnight ago.
The $117 is news in its own right. But there is a quieter piece of news from the same week, though, that probably matters more.
MCP Went Standard
MCP is really nothing else but a way for AI to talk to other systems (the DMS, the email, the practice management tool, the billing platform). And over the past few months it has become widely accepted as the standard protocol for that. Anthropic introduced it in late 2024, the other big AI providers picked it up, and now the legal tech vendors are wiring their systems for it too. NetDocuments shipped MCP connectivity to enterprise customers on April 1st, iManage followed in May, and Harvey and Legora have built their agentic workflows on top of it.
What this means for a boutique partner is that every system the firm runs should be able to speak MCP for any AI sitting on top of the firm (being it Claude, ChatGPT, or Gemini), to actually do anything useful with it. If a system doesn’t speak MCP, the AI can’t reach into it. The lawyer ends up shuttling context between the systems and the AI by hand, which kind of beats the whole purpose.
One practical thing you can do this week is to take stock. Go through the systems your firm pays for and check whether each one already has an MCP server, and if not, whether it’s on their roadmap. And if it’s not on the roadmap, that’s something to consider when you’re in the renewal conversation.
”Connect” as AI Strategy
Now, where does all of this fit in the firm’s bigger AI roadmap?
In the latest Institutional Moat article I split Stitch into two moves: Connect, which wires AI to systems the firm already pays for, and Extend, the agent layer built on top of what Connect makes reachable. The MCP is the protocol that makes the Connect possible and economically very attractive.
Essentially, there are four ways you can do a Connect today.
The first one is the native MCP connectors, the ones the vendors build and ship with their own product. Turn them on and they work. NetDocuments and iManage shipped theirs this spring, and more than 20 connectors for the legal industry were launched in May. The list keeps growing, and where one exists for a system your firm runs, this is the easiest way in.
The second one is Zapier or Make. Both have built their own MCP servers that expose thousands of business apps (9,000+ on Zapier, 3,000+ on Make). So if your firm uses a system that doesn’t have its own native MCP connector, chances are Zapier or Make do. You point the AI at their MCP endpoint and it can read from and write to whatever you’ve wired up. The catch is that in their case, the MCP server itself runs in their cloud, so the request data passes through them on its way to the AI. It’s worth knowing where the data goes.
The third one is n8n. Same idea as Zapier or Make, but n8n can be hosted on a machine the firm controls, so the data stays inside the firm’s infrastructure. This is the route to take when client work is sensitive enough that the PI carrier (or the partner) doesn’t want a third-party automation cloud sitting in the middle.
The fourth one is a custom MCP server. This one sounds intimidating, but it really isn’t. An MCP server is essentially connective tissue. It’s a small build, much lighter and less resource-heavy than what “building software” traditionally implies. It can be done in days instead of weeks, and the maintenance footprint is genuinely light. This is the route when the other three don’t reach the system you need to wire up, or when they don’t necessarily get the job done as well as you would like.
What $117 Redlining Is Built On
The firm behind it is Superlegal, founded by Noory Bechor and Ilan Admon, the same team that built LawGeex. They hold a Utah sandbox license to practice law as an entity rather than as individual lawyers, and they have rebuilt the delivery model around AI: the AI does the heavy lift, a licensed attorney signs off, and the trade association handles distribution.
The reason a firm like Superlegal would normally be hard to compete with is the head start. AI-native firms built their stacks from scratch - the AI infrastructure, the integration layer, the workflow plumbing. For a traditional firm to replicate that today would mean a multi-year R&D project nobody at scale wants to fund.
That’s the part that MCP addresses.
You don’t need an R&D project anymore to compete with these firms. And if you do the Connect, on top of the existing client history, relationships, and market reputation, it makes it really hard for new AI native startup firms to compete with you.
Legal AI in Action
🎬 What Did Cowork Do? Your Own Supervision Report On Team Plan
The data feed Cowork already sends out on the Team plan, how to catch it inside your own Microsoft tenant so the record stays where you control it, and the skill that turns it into a one-page supervision report you could hand to a client or your malpractice carrier.
🎙 Bleeding Edge at Home, Careful at the Firm
Pranav Anand is the EMEA digital forensics and incident response lead at Gravity Stack, with 17 years across DFIR and E-Discovery. Elgar Weijtmans is Head of Technology at HVG Law and a AI researcher at Legal Benchmarks, with a software engineering and lawyer background. Both run home labs to play with autonomous agents that aren’t ready for regulated legal yet. We talked through why that kind of experimentation teaches things a corporate pilot never will, where they keep humans in the loop on billable work, why firm wide knowledge management is finally starting to crack, and what it actually takes to set up experimentation inside a law firm.
Coming up!
🎙 Next Tuesday at 2pm CET!
Next week’s guests on Rok’s Legal AI Conversations is Kaichen Xu, founder and chief legal engineer of AI Lawyer Lab. Over the past 21 years he’s been partner at large firms, in-house GC at private equity and tech companies, and 13 years ago, founder of a legal tech startup.
We discuss the 30-50% automation threshold he’s modeled for moving from billable hours to fixed fees, why task by task automation is not the optimal approach for a firm trying to make money from AI, and what running a pilot practice (not a pilot project) actually looks like.
Each edition of Legal AI Brief brings practical lessons from firms using AI safely.