Harvey Opens Legal AI Connector Library With Document Write-Back

Harvey made its Connector Library generally available Oct. 21, giving legal teams a way to bring workplace information into its AI workspace and return completed work to document management systems. The library is designed to connect legal tasks with information held across tools used for documents, contracts, data rooms, email and chat.

Listed integrations include iManage, NetDocuments, Box, Google Drive, SharePoint/OneDrive, Outlook, Teams, Slack, Everlaw, Datasite and Docusign. Capabilities differ by connector. Harvey’s agent can recommend connectors for a task and provide citations linking its answers to source materials. The initial write-back capability covers document management systems, allowing work completed in Harvey to go back to those systems.

One example combines IT Slack threads, Jira outage tickets and contract data extracted by Docusign to assess whether a vendor agreement can be terminated. The workflow can then open a Docusign termination request, route it for approval and send the notice for signature. A separate example draws on case files, law firm work product and client instructions to prepare deposition outlines.

The library also supports legal intake. Requests can arrive through Slack or Jira, after which Harvey gathers context, routes the matter to an appropriate agent or lawyer, and replies in the original thread. The launch follows Harvey’s Legal Work Orchestration tool for recurring matters, which organizes work into stages and routes AI-assisted tasks through review.

Teams can also use custom Model Context Protocol (MCP) connectors to reach internal systems or third-party services outside the curated library. Workspace administrators enable these servers, and each user authenticates individually. The external server operator controls subsequent data processing. Some write, delete, send or delegated actions may run without a Harvey confirmation prompt; disconnecting a server also may not revoke an identity-provider token or remove data already submitted. These custom-server controls differ from the library’s connectors, which undergo security review and monitoring.

Harvey previously announced development of an MCP policy engine intended to govern tool access and agent actions.