The eDiscovery MCP server for Claude and ChatGPT.
Claira MCP is a remote Model Context Protocol server that turns your AI assistant into a document review teammate. Connect Claude or ChatGPT to Claira, then run privilege review, relevance review, objective coding, and PII detection over your evidence conversationally - without your documents ever leaving Claira.
An MCP server for legal document review, explained
The Model Context Protocol (MCP) is the open standard that lets AI assistants connect to outside systems. An MCP server exposes a product's capabilities as tools an assistant can call; an MCP client - Claude, ChatGPT, or any MCP-capable assistant - calls them on your behalf. It is the reason legal teams have started asking software vendors "do you have an MCP server?" the way they once asked "do you have an API?"
Claira MCP is that server for eDiscovery. It is a remote MCP server, authenticated with OAuth, that gives your assistant a legal-review toolset: check bulk review status, draft and test review prompts, launch privilege and relevance reviews, mine per-document results, generate cited chronologies and PII reports, keep case context current, and run field migrations. In short: an eDiscovery MCP integration that makes your general-purpose AI assistant fluent in document review.
Everything you'd do in Claira, asked for in plain language
Each capability is an MCP tool your assistant calls on your behalf. No new interface, no scripting, and no protocol knowledge required from the reviewer.
Monitor bulk review
Ask for review status, progress, failures, and skipped documents - and get a straight answer on what to do about each.
Launch privilege & relevance review
Describe the need. Claira drafts a review prompt from your case context, tests it on sample documents, and shows a cost preview before any bulk review starts.
Mine review results
Query per-document coding: which documents came back privileged, responsive, hot, or unsure - filtered, paginated, and ready to act on.
Generate cited insights
Chronologies, investigation narratives, relevance summaries, and PII reports, each claim citing the source document with a verification status.
Keep case context current
Paste in the amended complaint and review a section-by-section patch to the case context every review prompt is built from.
Run field migrations
Copy and map values between review fields before a production, with a compatibility check and an undoable one-document test first.
Works with the assistant your team already uses
Claira MCP is a standards-based remote MCP server, so it is not tied to one vendor. Add it once as a custom connector and it works everywhere that speaks the Model Context Protocol.
Claude
Claude Desktop, claude.ai, and Claude mobile. Add Claira as a remote MCP connector, authorize with OAuth, and start running eDiscovery from the chat.
ChatGPT
Add the Claira MCP server as a custom connector in ChatGPT and get the same legal review toolset, subject to the same server-side confirmations.
Any MCP-capable assistant
Because it implements the open Model Context Protocol over Streamable HTTP, any compliant MCP client can connect. One server, every assistant.
What puts eDiscovery inside the rest of your work
An API connects software to software. An MCP server connects your evidence to the assistant where the legal thinking actually happens. That is the difference that matters: because the same assistant reading your Claira review results is also the one drafting your correspondence, the chronology becomes the brief section, the results table becomes the client update, and the skipped-document list becomes tomorrow's plan. A document review MCP closes a loop no standalone eDiscovery platform can close on its own.
The questions your GC will ask
Claira MCP was designed around the legal data boundary first. Enterprise governance is the product, not an afterthought.
Results-first data boundary
By default the external assistant sees review results, statuses, and citations - never raw evidence text. Claira's region-resident models remain the only AI that reads your documents.
Admin-enabled, off by default
Organization administrators enable the MCP server and set the data tier per case. No connection exists until your firm deliberately creates one.
Preview and confirm on every action
Any tool call that spends budget or writes to a review field shows scope and cost first, then waits for a confirmation. Firms can require in-app approval above a threshold they set.
User-scoped permissions
The session carries your own case access and field security, enforced server-side. No path through the MCP can exceed what you could already do in the app.
Audit and attribution
Every tool call is logged with user, client, and timestamp. Reviews launched through an assistant are stamped in task history so you always know where they came from.
Regional endpoints
Region-resident MCP endpoints, no cross-region proxying, and the same SOC 2 control surface as the Claira app you already use.
Built for adversarial evidence
Opposing parties author the documents in your corpus, so prompt injection is not hypothetical in eDiscovery. Content crossing the boundary is returned as clearly delimited, provenance-tagged data rather than instructions; the assistant cannot invent a document population to act on; and nothing that spends or writes proceeds without a fresh human confirmation. An injected instruction has no unattended path to effect.
Be first in line
We're opening the eDiscovery MCP to a small group of law firms, corporate legal departments, and service providers. Tell us which assistant your team uses and how you'd put it to work.