Legal × Claude × Contract One MCP: Transforming Legal Operations in the AI Era with the Users Closest to the Work

Hello. I am Anna Odomari, a Product Manager (PdM) for the transaction management service "Contract One" at Sansan, Inc.
With Contract One surpassing 1 billion yen in ARR, I am pleased to contribute to this year's note relay series.

Recently, an article was published on the official Sansan note titled “Reducing Legal Review Time by 40%: What Changes When You Connect Claude to a 'Contract Database'?”.

It details how Sansan's legal team used the Claude × Contract One MCP server to reduce the time spent responding to legal inquiries by approximately 40%. This achievement was reached in about three months from the project kickoff.

Today, I would like to share what we, as Contract One PdMs, specifically did behind the scenes with the legal team.

Project ONE LEGAL

Contract One PdMs have been promoting the use of AI in product development for some time (Reference Article 1, Reference Article 2), but what we challenged ourselves with this time was an initiative to "become one team with the legal members, who are our closest users, and have them become active participants in product creation.".

The name of this project is Project ONE LEGAL. We named it to symbolize that Contract One and the legal department are becoming "one" as they evolve.

With PdMs providing support, legal members use Claude to reference contract data themselves and refine their judgment criteria into "Skills" (a collection of prompts and reference information for delegating specific tasks to Claude). This knowledge is then connected to the evolution of the product. We are running a cycle of
addressing both the reform of legal operations and the evolution of Contract One.

🚩 Goals set for Project ONE LEGAL

We set the above as an OKR for both the legal and product teams, designing it so we would pursue the same metrics together. We reorganized legal tasks into three groups based on the "nature of the judgment" and assigned a PdM and a designer to each group. We are working in a structure where product members can provide close support to each individual legal member.

What we did ①: Visualization of tasks and issues

The first thing we worked on was visualizing the tasks and issues of the legal members. We started by trying to inventory tasks and issues, but this proved quite difficult.
Legal members make numerous judgments on various types of consultations every day, but they don't necessarily verbalize each one. When they try to put into words what judgments they are actually making and what is taking up their time, it is difficult even for them to grasp.

The approach we took was to let AI handle the inventory first. We used Claude Code to analyze daily response records left in Notion and Slack inquiry threads. When we showed the results to the legal members, it naturally elicited realizations such as, "It's true that this is where time is being spent," or "I wasn't conscious of it, but I do this kind of work daily."

Building on that, we added qualitative information through interviews, such as "this involves this kind of judgment" or "this actually follows a set pattern."
Through this process, we were able to see "what is taking time" and "what are the repetitive tasks and judgments that can be standardized."

What we did ②: Preparing the environment

In parallel with the visualization of tasks, we proceeded with environment preparation.

▶︎ Distributing Claude accounts to legal members

Among several generative AI services, we chose Claude because we felt its strengths were highly compatible with legal work.

Word documents that go back and forth with revision history, long-running case threads... the unique nature of legal work, which requires accurately referencing various information and producing output, is very compatible with Claude's strengths (this is just my personal impression from comparing them at hand).
After using it a little, we started hearing voices from legal members saying, "A revolution is happening in legal work!!".

▶︎Connecting Claude with Contract One

In legal work, actions such as reviewing contracts, drafting documents, and responding to inquiries frequently involve referencing past or similar contracts. No matter how capable Claude is, if it cannot access the company's own contract data that needs to be referenced, it remains merely an "AI that makes legal work easier" rather than an "AI that makes decisions based on the company's context." Therefore, we connected Claude and Contract One using an MCP server. We created a state where
you can talk to the AI and it can search within Contract One to provide answers. Contract One contains not only the full text of contracts but also structured contract data, such as "parent-child relationships between contracts," "unique data managed by each company," and "corporate information of contract counterparties." By being able to access this, tasks that were originally searched for and interpreted manually can now be advanced with a single response from Claude.

▶︎Connecting with various services

Furthermore, we expanded the environment where Claude can access all the information necessary for legal work. The information needed for legal work is not limited to contract data. It includes the history of internal consensus building, past responses to security check sheets, and documents for each case. This information is scattered across multiple services. In Project ONE LEGAL, while organizing what information should be fed into the AI, we connected Claude not only to Contract One but also to various services such as Sansan, Notion, Box, Salesforce, and Slack.

What we did (3): Skill creation and operation

Once the work was visualized, we moved on to skill creation. The PdM worked closely with the team to create the first few drafts, and subsequent adjustments and new creations are handled by the legal members themselves. The people who know the work best are the ones doing it. As they operated it themselves, it evolved into a form where
each member cultivates skills that fit the field.

In all cases, Claude handles "research, verification, and drafting," while the legal members handle the decision-making itself. This division of roles has been incorporated into the form of Skills. These Skills have naturally become integrated into daily operations.

Results (1): Quantitative changes

The results of these initiatives lead to the figures in the official note introduced at the beginning. Legal inquiry response time reduced by approximately 40%. In fact, behind this reduction, the number of cases has increased by approximately 1.5 times. Even so, this result has been achieved. Looking at other data, we also saw interesting changes, such as a decrease in the number of back-and-forth exchanges and an increase in one-shot answers. This is just a figure at an interim point. Since skills are still being refined and there are areas we haven't touched yet, I believe we will easily reach our 50% goal in less than a month.

Results (2): Reduction of cognitive load

Along with quantitative changes, there was another important result: reduction of cognitive load. To be honest, even though I use Claude daily, I didn't really have the feeling that my "cognitive load had decreased." One day, while talking to a legal member, the topic came up that "since I started using Claude, my brain feels less burdened." When I asked other members, I found that many of them felt the same way. Tracing those voices revealed unique characteristics of legal work. In legal work,

referencing past contracts and exchanges
occurs daily whenever a decision is made. Moreover, it is not uncommon for a case to be idle for a week to several months while waiting for the other party's internal organization. Every time it resumed, it was necessary to re-read past correspondence records from scratch.

Both are changes born precisely because one can interact with Claude on a chat basis. Having the history summarized, brainstorming to organize thoughts—I feel that these interactive exchanges have greatly freed up the mental capacity that was previously used for "remembering, reassembling, and refining within oneself."

To the world as a Contract One MCP server

The knowledge accumulated in Project ONE LEGAL has been released as a Contract One MCP server so that it can be provided to customers after internal experimentation.

By connecting Claude and the Contract One MCP server, we have achieved great results in-house, but the form of legal work differs by company and industry, and the structure of challenges should also be different. We are currently taking on the challenge of working closely with each customer who feels challenged by contract review and information gathering from past contracts to achieve results together.

Conclusion — Why not create the future of legal work in the AI era together by utilizing Contract One?

Through this project, I have witnessed legal operations being transformed one after another as a quantitative result, and I have seen firsthand how legal team members are updating their own workflows while building their own skills. And I can see that there is still room for growth.
For my part, I feel that along with the evolution of AI technology, I want to not only evolve Contract One, but also the future of the legal field itself together with our customers.
If you read this article and thought, 'I'd like to try this at my company' or 'I'd like to hear more details,' please feel free to reach out!
I would be delighted if we could create the future of legal work together!