Legal AI MCP Connectors: Harvey, Legora, iManage

Over the past week, legal AI vendors announced more connections that let an assistant reach data it does not own. Legora and iManage said on October 16 that they plan to connect through iManage's MCP server. On October 19, LawSites reported that Midpage's case law is now available inside Harvey. The pattern matters more than any single link. Legora press release

What was announced

Legora and iManage (October 16). The companies announced plans to connect Legora with the iManage MCP Server, extending an existing API integration. For mutual customers, the stated goals are moving documents and updating document information through Legora workflows, and surfacing firm knowledge and past matters for organizations that use iManage Insight+. iManage permissions determine what each user can retrieve. The release gives no launch date. Legora

Harvey and Midpage (October 19). LawSites reports that Harvey customers can search Midpage's U.S. case law, statutes, regulations and agency materials from Harvey Assistant threads, and that the connector is in Harvey's Connector Library. Users need to be customers of both. The report does not mention price. This item rests on that one trade report; we did not find a Harvey or Midpage release. LawSites, October 19

Midpage elsewhere. Midpage's own site lists legal research and drafting integrations for Claude and ChatGPT and MCP access for legal products. Its Professional plan is listed at $100 a month. Midpage pricing, checked October 20

What MCP is doing here

MCP, the Model Context Protocol, is an open standard that lets an AI application call an outside source through one agreed connector instead of a custom integration for each pair. iManage launched its MCP server on May 14, 2026, and says access is "authenticated, permission-bound, and fully logged" and respects existing ethical walls. Those are vendor statements; the release names Harvey, Legora, ChatGPT, Claude and Microsoft Copilot as examples without saying which have been tested. iManage release, May 14

Read together, the announcements point to the same shape: the document store and the research database are separating from the assistant. The same Midpage library reaches Claude, ChatGPT and Harvey. The same iManage content can reach several assistants.

Why it matters for a firm buying AI

If your content stays in systems you control, switching assistants costs less. That is the practical meaning of "the model is not the system." It also weakens the case for paying one vendor for everything a seat is bundled with. The MCP guide for legal teams explains the mechanics.

A connector does not remove the risks. It moves the question from "which assistant" to "what can each assistant reach."

Questions to ask before you turn a connector on

1. What leaves our system when the assistant calls the connector, and what does the third party retain?

2. Does the connector use the end user's permissions, so ethical walls and matter restrictions still apply?

3. Where are calls logged, and can we export the log?

4. What do we pay for each side? A connector between two products can mean two subscriptions.

5. Who approves each new connector, and how do we remove one?

Put the answers in your vendor file. The pricing table shows which platforms publish a price today, and the pilot scorecard gives a place to record the results of a trial.

Operational information, not legal advice. Announcements checked October 20, 2026. Planned integrations are not released features, and vendor security claims were not independently verified.

Operational information, not legal advice. Announcements and vendor security claims are reported as vendor statements; one item rests on a single trade report, as noted. This update uses AI-assisted research and editing and is published under the byline of Manu Ayala.