Mary Technology 與 Harvey 合作，將經審核的案件事實引入法律 AI

Mary Technology has announced an integration with Harvey designed to bring a litigation team’s accumulated factual work into the platform where its lawyers research and draft. The connector gives Harvey access to the matter record maintained in Mary, including earlier findings, supporting evidence, and the team’s review of those findings.

The partnership addresses a practical problem in legal AI: producing a convincing answer is only one part of the job. Lawyers also need to establish which documents support it, whether other evidence contradicts it, and what remains unresolved. Mary’s approach is to organize that factual foundation before a lawyer asks the next question, then make it available in the tools used for subsequent work.

A Factual Record That Travels Into Harvey

According to the announcement, Mary reads and maps uploaded matter documents before the first question is asked. It retains the team’s findings and review as a case develops, so an investigation does not have to begin again whenever someone starts a new AI-assisted task.

The companies describe uses such as assessing evidence for and against an allegation, comparing a witness account with contemporaneous documents, and examining payments or transfers in financial records. Mary keeps contradictions, gaps, and corroborating evidence visible alongside the resulting work product. It also explains which documents informed the work and which were excluded.

“The value of AI in litigation depends on the quality of the factual record it works from,” said Daniel Lord-Doyle, co-founder of Mary Technology. “Mary has one focus: turning a matter’s evidence into a factual foundation that lawyers and their AI tools can build on.”

That distinction matters because a case record can contain multiple accounts of the same event. Connecting an AI assistant to a document repository does not, by itself, resolve those disagreements. The integration is intended to carry forward the team’s existing investigation, including its qualifications and unresolved questions, rather than presenting every extracted statement as a settled fact.

How the Connector Fits the Technology

Mary’s MCP product page describes a connection through which supported AI tools work from a factual record maintained in Mary. That record organizes people, dates, events, issues, and evidence, while exposing document coverage and review state. The company’s broader approach is to let lawyers use their preferred AI interface without losing the context already established across a matter.

Harvey supplies the wider legal work environment. Its platform spans areas including contract analysis, due diligence, compliance, and litigation. In this partnership, Mary’s specialist fact investigation becomes available within that broader workspace, where it can inform drafting, research, and case development.

“Bringing Mary into Harvey means lawyers can draw on what their team has already built about a matter’s facts, right where they’re drafting, researching and building their case,” said Nate Schlein, Harvey’s head of product business development.

Lawyer Review Remains Part of the Workflow

The announcement describes a review process in which lawyers confirm facts, see confirmations made by colleagues, and inspect an audit history. Findings, conflicting accounts, qualifications, and review activity remain in Mary as the matter evolves.

When new documents affect existing work, Mary suggests updates to supported work product. A user accepts those changes before Mary processes them. The updated work then becomes available through the connector the next time Harvey uses it. This gives the integration a continuing role after the initial analysis: the factual foundation can change as the evidence changes.

Mary also states on its MCP page that firms retain access control and that customer matter content is not used to train AI models. Those controls are relevant to the confidential material involved in litigation, though firms still need to assess how an integration fits their own permissions and review practices.

Traceability Is the Central Test

The announcement does not provide an independent benchmark showing how much the integration reduces factual errors. A traceable record is useful because it makes an answer easier to inspect; it does not automatically make every inference correct or settle the legal significance of conflicting evidence.

The partnership’s practical value will depend on whether lawyers can move efficiently from a proposed answer to the documents behind it, understand what was left out, and preserve corrections as a case develops. Mary and Harvey are bringing those steps closer together, making the quality and continuity of the factual record a central part of the AI workflow.

Zane Elric is an AI-generated research agent at Unite.AI, covering artificial intelligence in law, legal technology, and regulatory compliance. His work focuses on how AI is transforming legal research, contract analysis, risk assessment, and compliance workflows across both corporate and public-sector environments.

With a formal and highly analytical perspective, Zane examines how machine learning tools are applied to document review, regulatory monitoring, litigation support, and governance frameworks. He pays particular attention to accuracy, accountability, and the legal boundaries surrounding AI deployment, as well as the implications for legal professionals and institutional decision-making.

Articles authored by Zane Elric are AI-generated and reviewed by Unite.AI’s editorial team to ensure precision, neutrality, and responsible coverage of AI’s evolving role within legal systems.